StartLux-Decision-27B-Q8_0-GGUF

.gitattributes

@@ -0,0 +1,37 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text
+tokenizer.json filter=lfs diff=lfs merge=lfs -text
+StartLux-Decision-27B-Q8_0.gguf filter=lfs diff=lfs merge=lfs -text

LICENSE

@@ -0,0 +1,408 @@
+Attribution-NonCommercial 4.0 International
+
+=======================================================================
+
+Creative Commons Corporation ("Creative Commons") is not a law firm and
+does not provide legal services or legal advice. Distribution of
+Creative Commons public licenses does not create a lawyer-client or
+other relationship. Creative Commons makes its licenses and related
+information available on an "as-is" basis. Creative Commons gives no
+warranties regarding its licenses, any material licensed under their
+terms and conditions, or any related information. Creative Commons
+disclaims all liability for damages resulting from their use to the
+fullest extent possible.
+
+Using Creative Commons Public Licenses
+
+Creative Commons public licenses provide a standard set of terms and
+conditions that creators and other rights holders may use to share
+original works of authorship and other material subject to copyright
+and certain other rights specified in the public license below. The
+following considerations are for informational purposes only, are not
+exhaustive, and do not form part of our licenses.
+
+     Considerations for licensors: Our public licenses are
+     intended for use by those authorized to give the public
+     permission to use material in ways otherwise restricted by
+     copyright and certain other rights. Our licenses are
+     irrevocable. Licensors should read and understand the terms
+     and conditions of the license they choose before applying it.
+     Licensors should also secure all rights necessary before
+     applying our licenses so that the public can reuse the
+     material as expected. Licensors should clearly mark any
+     material not subject to the license. This includes other CC-
+     licensed material, or material used under an exception or
+     limitation to copyright. More considerations for licensors:
+    wiki.creativecommons.org/Considerations_for_licensors
+
+     Considerations for the public: By using one of our public
+     licenses, a licensor grants the public permission to use the
+     licensed material under specified terms and conditions. If
+     the licensor's permission is not necessary for any reason--for
+     example, because of any applicable exception or limitation to
+     copyright--then that use is not regulated by the license. Our
+     licenses grant only permissions under copyright and certain
+     other rights that a licensor has authority to grant. Use of
+     the licensed material may still be restricted for other
+     reasons, including because others have copyright or other
+     rights in the material. A licensor may make special requests,
+     such as asking that all changes be marked or described.
+     Although not required by our licenses, you are encouraged to
+     respect those requests where reasonable. More considerations
+     for the public:
+    wiki.creativecommons.org/Considerations_for_licensees
+
+=======================================================================
+
+Creative Commons Attribution-NonCommercial 4.0 International Public
+License
+
+By exercising the Licensed Rights (defined below), You accept and agree
+to be bound by the terms and conditions of this Creative Commons
+Attribution-NonCommercial 4.0 International Public License ("Public
+License"). To the extent this Public License may be interpreted as a
+contract, You are granted the Licensed Rights in consideration of Your
+acceptance of these terms and conditions, and the Licensor grants You
+such rights in consideration of benefits the Licensor receives from
+making the Licensed Material available under these terms and
+conditions.
+
+
+Section 1 -- Definitions.
+
+  a. Adapted Material means material subject to Copyright and Similar
+     Rights that is derived from or based upon the Licensed Material
+     and in which the Licensed Material is translated, altered,
+     arranged, transformed, or otherwise modified in a manner requiring
+     permission under the Copyright and Similar Rights held by the
+     Licensor. For purposes of this Public License, where the Licensed
+     Material is a musical work, performance, or sound recording,
+     Adapted Material is always produced where the Licensed Material is
+     synched in timed relation with a moving image.
+
+  b. Adapter's License means the license You apply to Your Copyright
+     and Similar Rights in Your contributions to Adapted Material in
+     accordance with the terms and conditions of this Public License.
+
+  c. Copyright and Similar Rights means copyright and/or similar rights
+     closely related to copyright including, without limitation,
+     performance, broadcast, sound recording, and Sui Generis Database
+     Rights, without regard to how the rights are labeled or
+     categorized. For purposes of this Public License, the rights
+     specified in Section 2(b)(1)-(2) are not Copyright and Similar
+     Rights.
+  d. Effective Technological Measures means those measures that, in the
+     absence of proper authority, may not be circumvented under laws
+     fulfilling obligations under Article 11 of the WIPO Copyright
+     Treaty adopted on December 20, 1996, and/or similar international
+     agreements.
+
+  e. Exceptions and Limitations means fair use, fair dealing, and/or
+     any other exception or limitation to Copyright and Similar Rights
+     that applies to Your use of the Licensed Material.
+
+  f. Licensed Material means the artistic or literary work, database,
+     or other material to which the Licensor applied this Public
+     License.
+
+  g. Licensed Rights means the rights granted to You subject to the
+     terms and conditions of this Public License, which are limited to
+     all Copyright and Similar Rights that apply to Your use of the
+     Licensed Material and that the Licensor has authority to license.
+
+  h. Licensor means the individual(s) or entity(ies) granting rights
+     under this Public License.
+
+  i. NonCommercial means not primarily intended for or directed towards
+     commercial advantage or monetary compensation. For purposes of
+     this Public License, the exchange of the Licensed Material for
+     other material subject to Copyright and Similar Rights by digital
+     file-sharing or similar means is NonCommercial provided there is
+     no payment of monetary compensation in connection with the
+     exchange.
+
+  j. Share means to provide material to the public by any means or
+     process that requires permission under the Licensed Rights, such
+     as reproduction, public display, public performance, distribution,
+     dissemination, communication, or importation, and to make material
+     available to the public including in ways that members of the
+     public may access the material from a place and at a time
+     individually chosen by them.
+
+  k. Sui Generis Database Rights means rights other than copyright
+     resulting from Directive 96/9/EC of the European Parliament and of
+     the Council of 11 March 1996 on the legal protection of databases,
+     as amended and/or succeeded, as well as other essentially
+     equivalent rights anywhere in the world.
+
+  l. You means the individual or entity exercising the Licensed Rights
+     under this Public License. Your has a corresponding meaning.
+
+
+Section 2 -- Scope.
+
+  a. License grant.
+
+       1. Subject to the terms and conditions of this Public License,
+          the Licensor hereby grants You a worldwide, royalty-free,
+          non-sublicensable, non-exclusive, irrevocable license to
+          exercise the Licensed Rights in the Licensed Material to:
+
+            a. reproduce and Share the Licensed Material, in whole or
+               in part, for NonCommercial purposes only; and
+
+            b. produce, reproduce, and Share Adapted Material for
+               NonCommercial purposes only.
+
+       2. Exceptions and Limitations. For the avoidance of doubt, where
+          Exceptions and Limitations apply to Your use, this Public
+          License does not apply, and You do not need to comply with
+          its terms and conditions.
+
+       3. Term. The term of this Public License is specified in Section
+          6(a).
+
+       4. Media and formats; technical modifications allowed. The
+          Licensor authorizes You to exercise the Licensed Rights in
+          all media and formats whether now known or hereafter created,
+          and to make technical modifications necessary to do so. The
+          Licensor waives and/or agrees not to assert any right or
+          authority to forbid You from making technical modifications
+          necessary to exercise the Licensed Rights, including
+          technical modifications necessary to circumvent Effective
+          Technological Measures. For purposes of this Public License,
+          simply making modifications authorized by this Section 2(a)
+          (4) never produces Adapted Material.
+
+       5. Downstream recipients.
+
+            a. Offer from the Licensor -- Licensed Material. Every
+               recipient of the Licensed Material automatically
+               receives an offer from the Licensor to exercise the
+               Licensed Rights under the terms and conditions of this
+               Public License.
+
+            b. No downstream restrictions. You may not offer or impose
+               any additional or different terms or conditions on, or
+               apply any Effective Technological Measures to, the
+               Licensed Material if doing so restricts exercise of the
+               Licensed Rights by any recipient of the Licensed
+               Material.
+
+       6. No endorsement. Nothing in this Public License constitutes or
+          may be construed as permission to assert or imply that You
+          are, or that Your use of the Licensed Material is, connected
+          with, or sponsored, endorsed, or granted official status by,
+          the Licensor or others designated to receive attribution as
+          provided in Section 3(a)(1)(A)(i).
+
+  b. Other rights.
+
+       1. Moral rights, such as the right of integrity, are not
+          licensed under this Public License, nor are publicity,
+          privacy, and/or other similar personality rights; however, to
+          the extent possible, the Licensor waives and/or agrees not to
+          assert any such rights held by the Licensor to the limited
+          extent necessary to allow You to exercise the Licensed
+          Rights, but not otherwise.
+
+       2. Patent and trademark rights are not licensed under this
+          Public License.
+
+       3. To the extent possible, the Licensor waives any right to
+          collect royalties from You for the exercise of the Licensed
+          Rights, whether directly or through a collecting society
+          under any voluntary or waivable statutory or compulsory
+          licensing scheme. In all other cases the Licensor expressly
+          reserves any right to collect such royalties, including when
+          the Licensed Material is used other than for NonCommercial
+          purposes.
+
+
+Section 3 -- License Conditions.
+
+Your exercise of the Licensed Rights is expressly made subject to the
+following conditions.
+
+  a. Attribution.
+
+       1. If You Share the Licensed Material (including in modified
+          form), You must:
+
+            a. retain the following if it is supplied by the Licensor
+               with the Licensed Material:
+
+                 i. identification of the creator(s) of the Licensed
+                    Material and any others designated to receive
+                    attribution, in any reasonable manner requested by
+                    the Licensor (including by pseudonym if
+                    designated);
+
+                ii. a copyright notice;
+
+               iii. a notice that refers to this Public License;
+
+                iv. a notice that refers to the disclaimer of
+                    warranties;
+
+                 v. a URI or hyperlink to the Licensed Material to the
+                    extent reasonably practicable;
+
+            b. indicate if You modified the Licensed Material and
+               retain an indication of any previous modifications; and
+
+            c. indicate the Licensed Material is licensed under this
+               Public License, and include the text of, or the URI or
+               hyperlink to, this Public License.
+
+       2. You may satisfy the conditions in Section 3(a)(1) in any
+          reasonable manner based on the medium, means, and context in
+          which You Share the Licensed Material. For example, it may be
+          reasonable to satisfy the conditions by providing a URI or
+          hyperlink to a resource that includes the required
+          information.
+
+       3. If requested by the Licensor, You must remove any of the
+          information required by Section 3(a)(1)(A) to the extent
+          reasonably practicable.
+
+       4. If You Share Adapted Material You produce, the Adapter's
+          License You apply must not prevent recipients of the Adapted
+          Material from complying with this Public License.
+
+
+Section 4 -- Sui Generis Database Rights.
+
+Where the Licensed Rights include Sui Generis Database Rights that
+apply to Your use of the Licensed Material:
+
+  a. for the avoidance of doubt, Section 2(a)(1) grants You the right
+     to extract, reuse, reproduce, and Share all or a substantial
+     portion of the contents of the database for NonCommercial purposes
+     only;
+
+  b. if You include all or a substantial portion of the database
+     contents in a database in which You have Sui Generis Database
+     Rights, then the database in which You have Sui Generis Database
+     Rights (but not its individual contents) is Adapted Material; and
+
+  c. You must comply with the conditions in Section 3(a) if You Share
+     all or a substantial portion of the contents of the database.
+
+For the avoidance of doubt, this Section 4 supplements and does not
+replace Your obligations under this Public License where the Licensed
+Rights include other Copyright and Similar Rights.
+
+
+Section 5 -- Disclaimer of Warranties and Limitation of Liability.
+
+  a. UNLESS OTHERWISE SEPARATELY UNDERTAKEN BY THE LICENSOR, TO THE
+     EXTENT POSSIBLE, THE LICENSOR OFFERS THE LICENSED MATERIAL AS-IS
+     AND AS-AVAILABLE, AND MAKES NO REPRESENTATIONS OR WARRANTIES OF
+     ANY KIND CONCERNING THE LICENSED MATERIAL, WHETHER EXPRESS,
+     IMPLIED, STATUTORY, OR OTHER. THIS INCLUDES, WITHOUT LIMITATION,
+     WARRANTIES OF TITLE, MERCHANTABILITY, FITNESS FOR A PARTICULAR
+     PURPOSE, NON-INFRINGEMENT, ABSENCE OF LATENT OR OTHER DEFECTS,
+     ACCURACY, OR THE PRESENCE OR ABSENCE OF ERRORS, WHETHER OR NOT
+     KNOWN OR DISCOVERABLE. WHERE DISCLAIMERS OF WARRANTIES ARE NOT
+     ALLOWED IN FULL OR IN PART, THIS DISCLAIMER MAY NOT APPLY TO YOU.
+
+  b. TO THE EXTENT POSSIBLE, IN NO EVENT WILL THE LICENSOR BE LIABLE
+     TO YOU ON ANY LEGAL THEORY (INCLUDING, WITHOUT LIMITATION,
+     NEGLIGENCE) OR OTHERWISE FOR ANY DIRECT, SPECIAL, INDIRECT,
+     INCIDENTAL, CONSEQUENTIAL, PUNITIVE, EXEMPLARY, OR OTHER LOSSES,
+     COSTS, EXPENSES, OR DAMAGES ARISING OUT OF THIS PUBLIC LICENSE OR
+     USE OF THE LICENSED MATERIAL, EVEN IF THE LICENSOR HAS BEEN
+     ADVISED OF THE POSSIBILITY OF SUCH LOSSES, COSTS, EXPENSES, OR
+     DAMAGES. WHERE A LIMITATION OF LIABILITY IS NOT ALLOWED IN FULL OR
+     IN PART, THIS LIMITATION MAY NOT APPLY TO YOU.
+
+  c. The disclaimer of warranties and limitation of liability provided
+     above shall be interpreted in a manner that, to the extent
+     possible, most closely approximates an absolute disclaimer and
+     waiver of all liability.
+
+
+Section 6 -- Term and Termination.
+
+  a. This Public License applies for the term of the Copyright and
+     Similar Rights licensed here. However, if You fail to comply with
+     this Public License, then Your rights under this Public License
+     terminate automatically.
+
+  b. Where Your right to use the Licensed Material has terminated under
+     Section 6(a), it reinstates:
+
+       1. automatically as of the date the violation is cured, provided
+          it is cured within 30 days of Your discovery of the
+          violation; or
+
+       2. upon express reinstatement by the Licensor.
+
+     For the avoidance of doubt, this Section 6(b) does not affect any
+     right the Licensor may have to seek remedies for Your violations
+     of this Public License.
+
+  c. For the avoidance of doubt, the Licensor may also offer the
+     Licensed Material under separate terms or conditions or stop
+     distributing the Licensed Material at any time; however, doing so
+     will not terminate this Public License.
+
+  d. Sections 1, 5, 6, 7, and 8 survive termination of this Public
+     License.
+
+
+Section 7 -- Other Terms and Conditions.
+
+  a. The Licensor shall not be bound by any additional or different
+     terms or conditions communicated by You unless expressly agreed.
+
+  b. Any arrangements, understandings, or agreements regarding the
+     Licensed Material not stated herein are separate from and
+     independent of the terms and conditions of this Public License.
+
+
+Section 8 -- Interpretation.
+
+  a. For the avoidance of doubt, this Public License does not, and
+     shall not be interpreted to, reduce, limit, restrict, or impose
+     conditions on any use of the Licensed Material that could lawfully
+     be made without permission under this Public License.
+
+  b. To the extent possible, if any provision of this Public License is
+     deemed unenforceable, it shall be automatically reformed to the
+     minimum extent necessary to make it enforceable. If the provision
+     cannot be reformed, it shall be severed from this Public License
+     without affecting the enforceability of the remaining terms and
+     conditions.
+
+  c. No term or condition of this Public License will be waived and no
+     failure to comply consented to unless expressly agreed to by the
+     Licensor.
+
+  d. Nothing in this Public License constitutes or may be interpreted
+     as a limitation upon, or waiver of, any privileges and immunities
+     that apply to the Licensor or You, including from the legal
+     processes of any jurisdiction or authority.
+
+=======================================================================
+
+Creative Commons is not a party to its public
+licenses. Notwithstanding, Creative Commons may elect to apply one of
+its public licenses to material it publishes and in those instances
+will be considered the “Licensor.” The text of the Creative Commons
+public licenses is dedicated to the public domain under the CC0 Public
+Domain Dedication. Except for the limited purpose of indicating that
+material is shared under a Creative Commons public license or as
+otherwise permitted by the Creative Commons policies published at
+creativecommons.org/policies, Creative Commons does not authorize the
+use of the trademark "Creative Commons" or any other trademark or logo
+of Creative Commons without its prior written consent including,
+without limitation, in connection with any unauthorized modifications
+to any of its public licenses or any other arrangements,
+understandings, or agreements concerning use of licensed material. For
+the avoidance of doubt, this paragraph does not form part of the
+public licenses.
+
+Creative Commons may be contacted at creativecommons.org.
+

NOTICE

@@ -0,0 +1,227 @@
+StartLux-Decision
+Copyright 2026 StartLux Labs
+
+Model weights
+-------------
+The StartLux-Decision model weights are licensed under the Creative Commons Attribution-NonCommercial 4.0
+International License (CC BY-NC 4.0); the full text is in LICENSE. You may use, share and adapt them for
+non-commercial purposes, with attribution. Commercial use requires a separate license from StartLux Labs:
+contact contact@startlux.com.
+
+The weights are a modified version of a model released under the Apache License, Version 2.0. That model's
+copyright notice is retained here:
+
+    Copyright 2026 Alibaba Cloud
+
+Files taken unchanged from that model, such as the tokenizer files, remain under the Apache License, Version 2.0.
+
+Inference code
+--------------
+The inference code in startlux_decision/ is licensed under the Apache License, Version 2.0.
+
+The full text of the Apache License, Version 2.0 follows.
+
+========================================================================
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.
+

README.md

@@ -0,0 +1,73 @@
+---
+license: cc-by-nc-4.0
+base_model: startlux-models/StartLux-Decision-27B
+base_model_relation: quantized
+tags:
+- gguf
+- llama.cpp
+- decision-model
+- typed-decisions
+---
+
+# StartLux-Decision-27B-Q8_0-GGUF
+
+[StartLux-Decision-27B](https://huggingface.co/startlux-models/StartLux-Decision-27B) in Q8_0 (8-bit) as a GGUF file for
+llama.cpp. The decision server in this repository turns it into the same typed decisions, with a probability for every
+option, as the original model.
+
+## Precisions
+
+<div align="center">
+
+| Repository | Bits | Size | Same decision as the original | JevBench public, of 231 | |
+|:---:|:---:|:---:|:---:|:---:|:---:|
+| **StartLux-Decision-27B-Q8_0-GGUF** (this repository) | 8-bit | 28.60 GB | 100.0% | 209 (90.5%) | recommended |
+| [StartLux-Decision-27B-Q4_K_M-GGUF](https://huggingface.co/startlux-models/StartLux-Decision-27B-Q4_K_M-GGUF) | 4-bit | 16.55 GB | 96.5% | 208 (90.0%) | smallest |
+| [StartLux-Decision-27B-BF16-GGUF](https://huggingface.co/startlux-models/StartLux-Decision-27B-BF16-GGUF) | 16-bit | 53.81 GB | 100.0% | 209 (90.5%) | the original weights, unchanged |
+| [StartLux-Decision-27B](https://huggingface.co/startlux-models/StartLux-Decision-27B) (original weights) | | | | 209 (90.5%) | for comparison |
+
+</div>
+
+"Same decision" is the share of the 231 public JevBench items on which the file picks the same answer as the original
+weights run through the `startlux_decision` package, with the same prompts, readout and temperatures. Q8_0 and Q4_K_M
+are llama.cpp's standard quantizations of BF16, without an importance matrix.
+
+## Download and run
+
+```bash
+hf download startlux-models/StartLux-Decision-27B-Q8_0-GGUF --local-dir StartLux-Decision-27B-Q8_0-GGUF
+cd StartLux-Decision-27B-Q8_0-GGUF
+pip install -r requirements.txt                     # transformers and torch; a CPU build of torch is enough
+
+# llama.cpp serves the weights; the decision server puts the prompt format, readout and calibration on top
+llama-server -m StartLux-Decision-27B-Q8_0.gguf -ngl 99 -c 16384 --parallel 4 --port 8081 &
+python -m startlux_decision.gguf_server --model-dir . --llama http://127.0.0.1:8081 --port 8090
+```
+
+`-ngl 99` puts every layer on the GPU (CUDA or Metal); leave it out on a CPU-only machine. llama.cpp has to be recent
+enough for this model (build b10454 or newer).
+
+Requests and responses use the TypeSafe `/v1/systemone` format:
+
+```bash
+curl -s localhost:8090/v1/systemone -H 'Content-Type: application/json' -d '{
+  "state": {"ticket": "I was charged twice for order #4411 and the app still shows it as unpaid."},
+  "questions": {
+    "team":   {"type": "choice", "instructions": "Which team should handle this ticket?",
+               "criteria": {"billing": "Payments, refunds and invoices",
+                            "shipping": "Delivery and tracking",
+                            "technical": "App, login and account problems"}},
+    "urgent": {"type": "noul", "instructions": "Should this ticket be answered today?"}
+  }
+}'
+```
+
+Plain chat with a GGUF file does not give these decisions: the option-letter readout and the per-type temperatures live
+in `gguf_server.py`, not in the weights. The file holds the text decoder only.
+
+## License
+
+The model weights are released under [CC BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/): free for
+research and other non-commercial use, with attribution. Commercial use requires a separate license from
+StartLux Labs; contact [contact@startlux.com](mailto:contact@startlux.com). The inference code in
+`startlux_decision/` is Apache-2.0. See [LICENSE](https://huggingface.co/startlux-models/StartLux-Decision-27B-Q8_0-GGUF/blob/main/LICENSE) and [NOTICE](https://huggingface.co/startlux-models/StartLux-Decision-27B-Q8_0-GGUF/blob/main/NOTICE).

StartLux-Decision-27B-Q8_0.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5868a8a2df103c63353adfbda6f57fec5b6393c9b8a1d97392855a0335001088
+size 28595763648

chat_template.jinja

@@ -0,0 +1,170 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- set reasoning_instructions = '' %}
+{%- if enable_thinking is undefined or enable_thinking is true %}
+    {%- set resolved_reasoning_effort = reasoning_effort|default('xhigh') %}
+    {%- if resolved_reasoning_effort not in ('xhigh', 'medium', 'low') %}
+        {{- raise_exception('Unexpected reasoning effort ' ~ reasoning_effort ~ '. Supported types are xhigh (default), medium, and low.') }}
+    {%- endif %}
+    {%- if resolved_reasoning_effort == 'xhigh' %}
+        {%- set reasoning_instructions = 'Reasoning effort is set to xhigh. Please think carefully through the task, validate key assumptions, consider plausible alternatives, and prioritize correctness, consistency, and clarity in the final answer.' %}
+    {%- elif resolved_reasoning_effort == 'low' %}
+        {%- set reasoning_instructions = 'Reasoning effort is set to low. Keep your thinking brief and focused, moving directly to the conclusion without unnecessary elaboration.' %}
+    {%- endif %}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {%- if reasoning_instructions %}
+        {{- reasoning_instructions + '\n\n' }}
+    {%- endif %}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '<|im_start|>system\n' + (reasoning_instructions + '\n\n' if reasoning_instructions else '')  + content + '<|im_end|>\n' }}
+        {%- elif reasoning_instructions %}
+            {{- '<|im_start|>system\n' + reasoning_instructions + '<|im_end|>\n' }}
+        {%- endif %}
+    {%- elif reasoning_instructions %}
+        {{- '<|im_start|>system\n' + reasoning_instructions + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if preserve_thinking is undefined or preserve_thinking is true or loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined and tool_call.arguments != '' %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | string if args_value is string else args_value | tojson | safe %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}

config.json

@@ -0,0 +1,140 @@
+{
+    "architectures": [
+        "Qwen3_5ForConditionalGeneration"
+    ],
+    "image_token_id": 248056,
+    "language_model_only": false,
+    "model_type": "qwen3_5",
+    "text_config": {
+        "attention_bias": false,
+        "attention_dropout": 0.0,
+        "attn_output_gate": true,
+        "bos_token_id": 248044,
+        "dtype": "bfloat16",
+        "eos_token_id": 248044,
+        "full_attention_interval": 4,
+        "head_dim": 256,
+        "hidden_act": "silu",
+        "hidden_size": 5120,
+        "initializer_range": 0.02,
+        "intermediate_size": 17408,
+        "layer_types": [
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention"
+        ],
+        "linear_conv_kernel_dim": 4,
+        "linear_key_head_dim": 128,
+        "linear_num_key_heads": 16,
+        "linear_num_value_heads": 48,
+        "linear_value_head_dim": 128,
+        "mamba_ssm_dtype": "float32",
+        "max_position_embeddings": 262144,
+        "model_type": "qwen3_5_text",
+        "mtp_num_hidden_layers": 1,
+        "mtp_use_dedicated_embeddings": false,
+        "num_attention_heads": 24,
+        "num_hidden_layers": 64,
+        "num_key_value_heads": 4,
+        "output_gate_type": "swish",
+        "pad_token_id": null,
+        "partial_rotary_factor": 0.25,
+        "rms_norm_eps": 1e-06,
+        "rope_parameters": {
+            "mrope_interleaved": true,
+            "mrope_section": [
+                11,
+                11,
+                10
+            ],
+            "partial_rotary_factor": 0.25,
+            "rope_theta": 10000000,
+            "rope_type": "default"
+        },
+        "tie_word_embeddings": false,
+        "use_cache": true,
+        "vocab_size": 248320
+    },
+    "tie_word_embeddings": false,
+    "transformers_version": "5.8.0.dev0",
+    "video_token_id": 248057,
+    "vision_config": {
+        "deepstack_visual_indexes": [],
+        "depth": 27,
+        "hidden_act": "gelu_pytorch_tanh",
+        "hidden_size": 1152,
+        "in_channels": 3,
+        "initializer_range": 0.02,
+        "intermediate_size": 4304,
+        "model_type": "qwen3_5",
+        "num_heads": 16,
+        "num_position_embeddings": 2304,
+        "out_hidden_size": 5120,
+        "patch_size": 16,
+        "spatial_merge_size": 2,
+        "temporal_patch_size": 2
+    },
+    "vision_end_token_id": 248054,
+    "vision_start_token_id": 248053
+}

decision_config.json

@@ -0,0 +1,42 @@
+{
+  "format": "StartLux-Decision-v1",
+  "letter_token_ids": [
+    32,
+    33,
+    34,
+    35,
+    36,
+    37,
+    38,
+    39,
+    40,
+    41,
+    42,
+    43,
+    44,
+    45,
+    46,
+    47,
+    48,
+    49,
+    50,
+    51,
+    52,
+    53,
+    54,
+    55,
+    56,
+    57
+  ],
+  "temperature_by_type": {
+    "noul": 1.421,
+    "choice": 1.2098,
+    "score": 1.2294
+  },
+  "max_options_per_pass": 26,
+  "wide_choice": {
+    "group": 25,
+    "keep": 3,
+    "residual": 0.001
+  }
+}

requirements.txt

@@ -0,0 +1,2 @@
+transformers
+torch

startlux_decision/__init__.py

@@ -0,0 +1,4 @@
+"""StartLux-Decision: typed decisions (choice, noul, score) with calibrated probabilities from one forward pass."""
+from .model import StartLuxDecision
+
+__all__ = ["StartLuxDecision"]

startlux_decision/check.py

@@ -0,0 +1,22 @@
+"""Check that the fast kernels will be used for a model, before starting a server on it.
+
+    python -m startlux_decision.check /path/to/StartLux-Decision-4B
+
+Exits with status 1 when flash-linear-attention or causal-conv1d is missing or not importable; transformers would then
+fall back to a plain torch path that is more than ten times slower.
+"""
+import sys
+
+from .model import fast_kernels_active
+
+
+def main():
+    if len(sys.argv) != 2:
+        sys.exit(__doc__.strip())
+    ok = fast_kernels_active(sys.argv[1])
+    print("fast kernels: " + ("active" if ok else "NOT active, pip install flash-linear-attention causal-conv1d"))
+    sys.exit(0 if ok else 1)
+
+
+if __name__ == "__main__":
+    main()

startlux_decision/gguf_server.py

@@ -0,0 +1,108 @@
+"""A /v1/systemone server for a StartLux-Decision GGUF file run by llama.cpp.
+
+    llama-server -m StartLux-Decision-4B-Q8_0.gguf -ngl 99 -c 16384 --parallel 4 --port 8081
+    python -m startlux_decision.gguf_server --model-dir StartLux-Decision-4B-GGUF --llama http://127.0.0.1:8081 --port 8090
+
+It is StartLuxDecision with only the forward pass replaced by a llama-server call: prompt rendering, the option-letter
+readout, the per-type temperatures and wide choices are the package's own code, so the answers match the original
+model question by question up to the numerics of the GGUF file.  MODEL_DIR holds the tokenizer files, config.json and
+decision_config.json (the GGUF repositories ship them next to the .gguf files).  llama-server returns the
+log-probabilities of the whole vocabulary at the answer position; dividing them by the temperature and normalising over
+the listed options is the same as doing it on the logits."""
+import argparse
+import json
+import os
+import urllib.request
+from concurrent.futures import ThreadPoolExecutor
+from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
+
+import torch
+
+from . import jevfmt as J
+from .model import StartLuxDecision
+
+
+class GGUFDecision(StartLuxDecision):
+    def __init__(self, path, llama, workers):                     # no torch model: llama-server runs the weights
+        from transformers import AutoTokenizer
+        cfg = json.load(open(os.path.join(path, "decision_config.json")))
+        self.tok = AutoTokenizer.from_pretrained(path)
+        self.letters = J.check_tokenizer(self.tok)
+        if self.letters != cfg["letter_token_ids"]:
+            raise ValueError("tokenizer letter ids differ from decision_config.json")
+        self.temperature = {k: float(v) for k, v in cfg["temperature_by_type"].items()}
+        wide = cfg.get("wide_choice", {})
+        self.group, self.keep, self.residual = wide.get("group", 25), wide.get("keep", 3), wide.get("residual", 1e-3)
+        self.max_length, self.graphs = 65536, {}
+        mc = json.load(open(os.path.join(path, "config.json")))
+        self.vocab = int(mc.get("text_config", mc).get("vocab_size", 262144))
+        self.url = llama.rstrip("/") + "/completion"
+        self.pool = ThreadPoolExecutor(workers)
+
+    def _letter_logprobs(self, ids, count):
+        """Next-token log-probabilities of the option letters at the last prompt position (full-vocabulary softmax;
+        dividing by the temperature and normalising over the options afterwards equals doing it on the logits)."""
+        targets = self.letters[:count]
+        for n in (256, self.vocab):
+            body = json.dumps({"prompt": ids, "n_predict": 1, "temperature": -1, "n_probs": n, "cache_prompt": False}).encode()
+            with urllib.request.urlopen(urllib.request.Request(self.url, body, {"Content-Type": "application/json"}), timeout=900) as r:
+                out = json.loads(r.read())
+            rows = out.get("completion_probabilities", out.get("probs"))
+            got = {t["id"]: t["logprob"] for t in rows[0]["top_logprobs"]}
+            if all(t in got for t in targets):
+                return [got[t] for t in targets]
+        raise RuntimeError("option letters missing from llama-server's log-probabilities")
+
+    def _logits(self, rows):
+        jobs = []
+        for r in rows:
+            order = [o["id"] for o in r["options"]]
+            ids, _ = J.render_ids(r, self.tok, order, max_length=self.max_length)
+            jobs.append((ids, len(order)))
+        outs = list(self.pool.map(lambda j: self._letter_logprobs(*j), jobs))
+        return [torch.tensor(o, dtype=torch.float32) for o in outs], sum(len(i) for i, _ in jobs)
+
+
+def main():
+    ap = argparse.ArgumentParser()
+    ap.add_argument("--model-dir", required=True)
+    ap.add_argument("--llama", required=True)
+    ap.add_argument("--port", type=int, required=True)
+    ap.add_argument("--workers", type=int, default=16)
+    ap.add_argument("--name", help="model name reported in responses (default: the GGUF directory name)")
+    a = ap.parse_args()
+    engine = GGUFDecision(a.model_dir, a.llama, a.workers)
+    a.name = a.name or os.path.basename(os.path.abspath(a.model_dir))
+
+    class Handler(BaseHTTPRequestHandler):
+        def _send(self, code, obj):
+            data = json.dumps(obj).encode()
+            self.send_response(code)
+            self.send_header("Content-Type", "application/json")
+            self.send_header("Content-Length", str(len(data)))
+            self.end_headers()
+            self.wfile.write(data)
+
+        def do_GET(self):
+            self._send(200, {"status": "ok", "model": a.name})
+
+        def do_POST(self):
+            try:
+                req = json.loads(self.rfile.read(int(self.headers.get("Content-Length", 0))))
+                answers, usage = engine.decide(req["state"], req["questions"])
+                self._send(200, {"answers": answers, "usage": usage, "model": a.name})
+            except Exception as e:                                 # surfaced to the client; suites.py stops on it
+                self._send(500, {"error": repr(e)})
+
+        def log_message(self, *args):
+            pass
+
+    class Server(ThreadingHTTPServer):
+        request_queue_size = 1024                           # the default backlog of 5 resets connections under load
+        daemon_threads = True
+
+    Server(("127.0.0.1", a.port), Handler).serve_forever()
+
+
+if __name__ == "__main__":
+    main()

startlux_decision/jevfmt.py

@@ -0,0 +1,188 @@
+"""Prompt rendering for StartLux-Decision.
+
+A question about a state is turned into one chat prompt: a fixed system line, then an Evidence / Question / Options
+block with lettered options, and the thinking-off assistant prefix.  The answer is read from the next-token logits of
+the option letters at the last prompt position, so nothing is generated.
+
+Rules worth knowing when you build requests by hand: an option without a description is shown by its id alone, yes/no
+questions are shown as yes / no, choice ids that are bare letters or numbers are hidden (a line such as "A) C: Paris"
+would make the answer letter ambiguous).  StartLux-Decision shows options in the order the request lists them; score levels
+are always lowest first.
+"""
+import hashlib
+import json
+import re
+import string
+
+VERSION = "jev_render_v3"
+SYSTEM = "Apply the criterion to the evidence. Choose exactly one listed option. Answer with its letter only."
+LETTERS = string.ascii_uppercase
+TYPES = ("choice", "score", "noul")
+MAX_OPTIONS = 26
+MAX_LEVELS = 10
+_BARE = re.compile(r"^(?:\(?[A-Za-z][\).]?|\(?\d{1,3}[\).]?|option[_ ]?\d{1,3}|opt[_ ]?\d{1,3})$", re.I)
+ANNOTATE_MIN = 8
+
+
+def short_hash(text, n=16):
+    return hashlib.sha1(text.encode("utf-8", "surrogatepass")).hexdigest()[:n]
+
+
+def annotate_indices(x, min_len=ANNOTATE_MIN):
+    """Long arrays carry their element index so that items can be referred to by position."""
+    if isinstance(x, list):
+        if len(x) >= min_len:
+            return [({"_index": i, **annotate_indices(v, min_len)} if isinstance(v, dict)
+                     else {"_index": i, "value": annotate_indices(v, min_len)}) for i, v in enumerate(x)]
+        return [annotate_indices(v, min_len) for v in x]
+    if isinstance(x, dict):
+        return {k: annotate_indices(v, min_len) for k, v in x.items()}
+    return x
+
+
+def state_text(state):
+    if isinstance(state, str):
+        return state
+    if state is None or state == {} or state == []:
+        return ""
+    return json.dumps(annotate_indices(state), ensure_ascii=False)
+
+
+def value_text(v):
+    if v is None:
+        return None
+    s = v if isinstance(v, str) else json.dumps(v, ensure_ascii=False)
+    s = s.strip()
+    return s or None
+
+
+def _norm(s):
+    return re.sub(r"[\s_\-]+", " ", str(s)).strip().lower()
+
+
+def validate(row):
+    t = row.get("type")
+    if t not in TYPES:
+        raise ValueError(f"unknown type {t!r}")
+    if not isinstance(row.get("state"), str):
+        raise ValueError("state must be a string")
+    if not isinstance(row.get("instructions"), str) or not row["instructions"].strip():
+        raise ValueError("instructions required")
+    opts = row.get("options")
+    if not isinstance(opts, list) or not 2 <= len(opts) <= MAX_OPTIONS:
+        raise ValueError(f"2..{MAX_OPTIONS} options required, got {len(opts) if isinstance(opts, list) else opts!r}")
+    ids = []
+    for o in opts:
+        i = o.get("id")
+        if not isinstance(i, str) or not i.strip() or "\n" in i or "\r" in i or len(i) > 200:
+            raise ValueError(f"bad option id {i!r}")
+        c = o.get("criterion")
+        if c is not None and not isinstance(c, str):
+            raise ValueError("criterion must be a string or None")
+        ids.append(i)
+    if len(set(ids)) != len(ids):
+        raise ValueError("duplicate option id")
+    if t == "noul" and set(ids) != {"true", "false"}:
+        raise ValueError("noul ids must be true/false")
+    if t == "score" and not 2 <= len(ids) <= MAX_LEVELS:
+        raise ValueError(f"score needs 2..{MAX_LEVELS} levels")
+    return ids
+
+
+def option_lines(row, order):
+    crit = {o["id"]: o.get("criterion") for o in row["options"]}
+    t = row["type"]
+    hide = t == "choice" and all(_BARE.match(i) for i in order) and all(crit[i] for i in order)
+    lines = []
+    for k, i in enumerate(order):
+        name = ("yes" if i == "true" else "no") if t == "noul" else i
+        c = crit[i]
+        if hide:
+            body = c
+        elif c is None or not c.strip() or _norm(c) == _norm(name):
+            body = name
+        else:
+            body = f"{name}: {c}"
+        lines.append(f"{LETTERS[k]}) {body}")
+    return lines
+
+
+def messages(row, order=None):
+    validate(row)
+    order = [o["id"] for o in row["options"]] if order is None else list(order)
+    if sorted(order) != sorted(o["id"] for o in row["options"]):
+        raise ValueError("order must be a permutation of option ids")
+    state = row["state"] if row["state"].strip() else "(none)"
+    content = ("Evidence:\n" + state + "\n\nQuestion: " + row["instructions"].strip() + "\nOptions:\n"
+               + "\n".join(option_lines(row, order)))
+    return [{"role": "system", "content": SYSTEM}, {"role": "user", "content": content}], order
+
+
+THINK_OFF_SUFFIX = "<think>\n\n</think>\n\n"
+
+
+def check_tokenizer(tokenizer):
+    """Letter ids and the native thinking-off prefix; returns the 26 letter token ids."""
+    probe, _ = messages({"type": "noul", "state": "s", "instructions": "q",
+                         "options": [{"id": "true", "criterion": None}, {"id": "false", "criterion": None}]})
+    text = tokenizer.apply_chat_template(probe, tokenize=False, add_generation_prompt=True, enable_thinking=False)
+    if not text.endswith(THINK_OFF_SUFFIX):
+        raise ValueError("tokenizer chat template lacks the thinking-off assistant prefix")
+    ids = tokenizer.encode(text, add_special_tokens=False)
+    letters = []
+    for L in LETTERS:
+        t = tokenizer.encode(L, add_special_tokens=False)
+        if len(t) != 1 or tokenizer.encode(text + L, add_special_tokens=False) != ids + t:
+            raise ValueError("letter is not a separate single token after the prefix: " + L)
+        letters.append(t[0])
+    if len(set(letters)) != 26:
+        raise ValueError("letter ids not distinct")
+    return letters
+
+
+def render_ids(row, tokenizer, order=None, max_length=12288):
+    """-> (input_ids, order). The answer letter is read at position len(input_ids) - 1."""
+    msgs, order = messages(row, order)
+    text = tokenizer.apply_chat_template(msgs, tokenize=False, add_generation_prompt=True, enable_thinking=False)
+    ids = tokenizer.encode(text, add_special_tokens=False)
+    if len(ids) > max_length:
+        raise ValueError(f"length {len(ids)} > {max_length}")
+    return ids, order
+
+
+# ---------------------------------------------------------------- /v1/systemone requests
+
+def score_keys(crit):
+    """Level order of a legend-form Score: numeric keys ascending, otherwise as given."""
+    try:
+        return sorted(crit, key=lambda k: float(k))
+    except (TypeError, ValueError):
+        return list(crit)
+
+
+def from_systemone(state, spec, qid="q"):
+    """One /v1/systemone question spec -> the record that gets rendered. Score level ids are "0".."n-1"."""
+    t = spec.get("type", "choice")
+    t = "noul" if t == "bool" else t
+    crit = spec.get("criteria", spec.get("options"))
+    ins = value_text(spec.get("instructions", spec.get("question"))) or ""
+    if t == "choice":
+        if isinstance(crit, (list, tuple)):
+            crit = {str(c): None for c in crit}
+        opts = [{"id": str(k), "criterion": value_text(v)} for k, v in crit.items()]
+    elif t == "score":
+        if isinstance(crit, dict):
+            crit = [crit[k] for k in score_keys(crit)]
+        opts = [{"id": str(i), "criterion": value_text(c)} for i, c in enumerate(crit)]
+    elif t == "noul":
+        c = crit if isinstance(crit, dict) else {}
+        tr, fa = c.get("true", c.get(True)), c.get("false", c.get(False))
+        opts = [{"id": "true", "criterion": value_text(tr)}, {"id": "false", "criterion": value_text(fa)}]
+        if not ins:
+            ins = "Which answer fits the evidence?"
+    else:
+        raise ValueError(f"unknown question type {t!r}")
+    row = {"id": qid, "type": t, "state": state_text(state), "instructions": ins, "options": opts}
+    if t == "score":
+        row["ordered"] = True
+    return row

startlux_decision/model.py

@@ -0,0 +1,293 @@
+"""StartLux-Decision inference.
+
+Each question is rendered as one prompt (startlux_decision/jevfmt.py), the model runs one forward pass, and the
+answer is read from the next-token logits of the option letters at the last prompt position, divided by the
+temperature of the question type and normalised over the listed options.  Nothing is generated.
+
+All questions of one request run in one forward pass, one row per question.  On CUDA the pass is a recorded graph:
+one per padded input length (128 ... 4096 tokens) for a single question, and one per (question count, length) for two
+to four questions of up to 1024 tokens each.  Replaying a graph removes the per-layer launch overhead (the approach of
+the JevK5 runtime, Apache-2.0).  Inputs are right-padded; every layer is causal and rows never mix, so padding after the
+last prompt token never reaches the position that is read.  Other requests run eagerly as one padded batch (inputs
+padded to one of PAD_LENGTHS); decide_batch() batches questions across many requests.
+"""
+import importlib
+import json
+import os
+import warnings
+
+import torch
+
+from . import jevfmt as J
+
+__all__ = ["StartLuxDecision", "load_model", "fast_kernels_active"]
+
+GRAPH_LENGTHS = (128, 192, 256, 320, 384, 512, 640, 768, 1024, 1536, 2048, 3072, 4096)
+GRAPH_ROWS = (1, 2, 3, 4)          # questions per request replayed as one graph
+MULTI_ROW_MAX_LENGTH = 1024        # longest prompt for the multi-question graphs (rows x length <= 4096 tokens)
+# Batched inputs are padded to one of these lengths.  The linear-attention kernels are compiled once per sequence length
+# (about a second each), so padding to exact lengths would recompile for almost every batch.  Above 256 tokens a step
+# adds at most 25% padding.
+PAD_LENGTHS = (128, 192, 256, 320, 384, 448, 512, 640, 768, 896, 1024, 1280, 1536, 1792, 2048, 2560, 3072, 3584, 4096,
+               5120, 6144, 7168, 8192, 10240, 12288, 14336, 16384, 20480, 24576, 28672, 32768, 40960, 49152, 57344, 65536)
+
+
+def padded_length(n):
+    """The length a batch whose longest input has n tokens is padded to."""
+    for length in PAD_LENGTHS:
+        if n <= length:
+            return length
+    return -(-n // 8192) * 8192
+
+
+def fast_kernels_active(path):
+    """True when transformers will use the fla / causal-conv1d kernels for the linear-attention layers of the model
+    in `path`."""
+    from transformers import AutoConfig
+
+    kind = AutoConfig.from_pretrained(path).model_type
+    try:
+        modeling = importlib.import_module(f"transformers.models.{kind}.modeling_{kind}")
+    except ImportError:
+        return False
+    return bool(getattr(modeling, "is_fast_path_available", False))
+
+
+def load_model(path, device):
+    """The checkpoint's own transformers class in bf16 on `device`, and its text decoder (the stack without the
+    output head; checkpoints that also carry other towers keep the text decoder under .language_model)."""
+    import transformers
+
+    arch = transformers.AutoConfig.from_pretrained(path).architectures[0]
+    model = getattr(transformers, arch).from_pretrained(path, dtype=torch.bfloat16, device_map={"": device})
+    return model, getattr(model.model, "language_model", model.model)
+
+
+class StartLuxDecision:
+    """decide(state, questions) -> (answers, usage), answers in the TypeSafe /v1/systemone format."""
+
+    def __init__(self, path, device=None, max_length=65536, max_batch_tokens=65536, graphs=True):
+        from transformers import AutoTokenizer
+
+        if not os.path.isdir(path):
+            raise FileNotFoundError(f"{path}: expected a local StartLux-Decision directory (weights are shared separately)")
+        cfg = json.load(open(os.path.join(path, "decision_config.json")))
+        self.tok = AutoTokenizer.from_pretrained(path)
+        self.letters = J.check_tokenizer(self.tok)
+        if self.letters != cfg["letter_token_ids"]:
+            raise ValueError("tokenizer letter ids differ from decision_config.json")
+        self.temperature = {k: float(v) for k, v in cfg["temperature_by_type"].items()}
+        wide = cfg.get("wide_choice", {})
+        self.group, self.keep, self.residual = wide.get("group", 25), wide.get("keep", 3), wide.get("residual", 1e-3)
+        self.device = torch.device(device or ("cuda" if torch.cuda.is_available() else "cpu"))
+        self.fast_kernels = fast_kernels_active(path)
+        if self.device.type == "cuda" and not self.fast_kernels:
+            msg = ("flash-linear-attention and causal-conv1d are not active, so transformers would run the plain torch "
+                   "path of the linear-attention layers (>10x slower). pip install flash-linear-attention causal-conv1d, "
+                   "or set STARTLUX_ALLOW_SLOW=1 to run anyway.")
+            if os.environ.get("STARTLUX_ALLOW_SLOW") != "1":
+                raise RuntimeError(msg)
+            warnings.warn(msg)
+        model, body = load_model(path, self.device)
+        self.body = body.eval()
+        head = model.get_output_embeddings().weight
+        self.letter_rows = head.index_select(0, torch.tensor(self.letters, device=head.device)).float()
+        del model                       # only the decoder and the letter rows of the output head are used
+        self.pad = self.tok.pad_token_id
+        self.max_length, self.max_batch_tokens = int(max_length), int(max_batch_tokens)
+        self.graphs = {}
+        if graphs and self.device.type == "cuda" and os.environ.get("STARTLUX_GRAPHS", "1") != "0":
+            self._capture()
+
+    # ---- CUDA-graph path (the questions of one request as rows, right-padded, no attention mask)
+    def _slot_logits(self, ids, last):
+        with torch.autocast(self.device.type, dtype=torch.bfloat16):
+            hidden = self.body(input_ids=ids, use_cache=False, return_dict=True).last_hidden_state
+        h = hidden[torch.arange(ids.shape[0], device=ids.device), last].float()
+        return h @ self.letter_rows.T
+
+    @torch.inference_mode()
+    def _capture(self):
+        shapes = [(b, n) for b in GRAPH_ROWS for n in GRAPH_LENGTHS if b == 1 or n <= MULTI_ROW_MAX_LENGTH]
+        pool = None                                 # one memory pool for every graph, sized by the largest
+        for b, n in sorted(shapes, key=lambda s: (-s[0] * s[1], -s[1])):
+            ids = torch.full((b, n), self.pad, dtype=torch.long, device=self.device)
+            last = torch.full((b,), n - 1, dtype=torch.long, device=self.device)
+            stream = torch.cuda.Stream(device=self.device)
+            stream.wait_stream(torch.cuda.current_stream(self.device))
+            with torch.cuda.stream(stream):
+                for _ in range(3):                  # autotune and warm every kernel outside the capture
+                    self._slot_logits(ids, last)
+            torch.cuda.current_stream(self.device).wait_stream(stream)
+            graph = torch.cuda.CUDAGraph()
+            with torch.cuda.graph(graph, pool=pool):
+                out = self._slot_logits(ids, last)
+            pool = graph.pool()
+            self.graphs[(b, n)] = (graph, ids, last, out)
+
+    def _graph_length(self, enc):
+        """The recorded length for these rows, or None when no graph fits."""
+        longest = max(len(t) for _, t, _ in enc)
+        fits = [n for b, n in self.graphs if b == len(enc) and n >= longest]
+        return min(fits) if fits else None
+
+    @torch.inference_mode()
+    def _graph_logits(self, enc, n):
+        graph, static_ids, last, out = self.graphs[(len(enc), n)]
+        ids = torch.full((len(enc), n), self.pad, dtype=torch.long)
+        for j, (_, t, _) in enumerate(enc):
+            ids[j, :len(t)] = torch.tensor(t)
+        static_ids.copy_(ids)
+        last.copy_(torch.tensor([len(t) - 1 for _, t, _ in enc]))
+        graph.replay()
+        res = out.cpu()
+        return [res[j, :c].clone() for j, (_, _, c) in enumerate(enc)]
+
+    # ---- readout
+    @torch.no_grad()
+    def _logits(self, rows):
+        """rows: rendered records (<= 26 options) -> (letter logits per row in option order, prompt tokens)."""
+        enc = []
+        for i, r in enumerate(rows):
+            order = [o["id"] for o in r["options"]]
+            ids, _ = J.render_ids(r, self.tok, order, max_length=self.max_length)
+            enc.append((i, ids, len(order)))
+        out, tokens = [None] * len(rows), 0
+        n = self._graph_length(enc) if self.graphs and enc else None
+        if n is not None:                           # every question of the request in one replay
+            for (i, t, _), z in zip(enc, self._graph_logits(enc, n)):
+                out[i] = z
+                tokens += len(t)
+            return out, tokens
+        enc.sort(key=lambda x: -len(x[1]))
+        start = 0
+        while start < len(enc):
+            longest = padded_length(len(enc[start][1]))
+            n = max(1, min(len(enc) - start, self.max_batch_tokens // longest))
+            batch = enc[start:start + n]
+            start += n
+            ids = torch.full((len(batch), longest), self.pad, dtype=torch.long)
+            attn = torch.zeros_like(ids)
+            for j, (_, t, _) in enumerate(batch):
+                ids[j, :len(t)] = torch.tensor(t)
+                attn[j, :len(t)] = 1
+                tokens += len(t)
+            ids, attn = ids.to(self.device), attn.to(self.device)
+            with torch.autocast(self.device.type, dtype=torch.bfloat16):
+                hidden = self.body(input_ids=ids, attention_mask=attn, use_cache=False, return_dict=True).last_hidden_state
+            pos = attn.sum(1) - 1
+            h = hidden[torch.arange(len(batch), device=self.device), pos].float()
+            logits = h @ self.letter_rows.T
+            for j, (i, _, c) in enumerate(batch):
+                out[i] = logits[j, :c].cpu()
+        return out, tokens
+
+    def _probs(self, rows):
+        logits, tokens = self._logits(rows)
+        return [torch.softmax(z / self.temperature.get(r["type"], 1.0), -1).tolist() for r, z in zip(rows, logits)], tokens
+
+    def self_test(self, state, questions):
+        """Largest |p_graph - p_eager| over the questions of one request (0.0 when no graphs were recorded)."""
+        if not self.graphs:
+            return 0.0
+        rows = [J.from_systemone(state, q) for q in questions.values()]
+        graphed, _ = self._probs(rows)
+        graphs, self.graphs = self.graphs, {}
+        try:
+            eager, _ = self._probs(rows)
+        finally:
+            self.graphs = graphs
+        return max(abs(a - b) for pg, pe in zip(graphed, eager) for a, b in zip(pg, pe))
+
+    def _wide(self, state, spec):
+        """Choice lists over 26 options: near-equal groups, the top `keep` of each group go to a final round."""
+        keys = list(spec["criteria"])
+        n_groups = -(-len(keys) // self.group)
+        size, extra = divmod(len(keys), n_groups)
+        groups, start = [], 0
+        for g in range(n_groups):
+            end = start + size + (1 if g < extra else 0)
+            groups.append(keys[start:end])
+            start = end
+        rows = [J.from_systemone(state, dict(spec, criteria={k: spec["criteria"][k] for k in g})) for g in groups]
+        first, tokens = self._probs(rows)
+        first_p = {k: p for g, ps in zip(groups, first) for k, p in zip(g, ps)}
+        finalists = [k for g, ps in zip(groups, first) for k, _ in sorted(zip(g, ps), key=lambda x: -x[1])[:self.keep]]
+        fin_spec = dict(spec, criteria={k: spec["criteria"][k] for k in finalists})
+        if len(finalists) > J.MAX_OPTIONS:
+            final_p, more = self._wide(state, fin_spec)
+        else:
+            ps, more = self._probs([J.from_systemone(state, fin_spec)])
+            final_p = dict(zip(finalists, ps[0]))
+        rest = [k for k in keys if k not in set(finalists)]
+        mass = sum(first_p[k] for k in rest) or 1.0
+        probs = {k: final_p[k] * (1 - self.residual) for k in finalists}
+        probs.update({k: self.residual * first_p[k] / mass for k in rest})
+        z = sum(probs.values())
+        return {k: v / z for k, v in probs.items()}, tokens + more
+
+    # ---- public API
+    @staticmethod
+    def _answer(row, p, question):
+        ids = [o["id"] for o in row["options"]]
+        if row["type"] == "noul":
+            return {"type": "noul", "noul": p[ids.index("true")]}
+        if row["type"] == "score":
+            levels = question.get("criteria") or []
+            levels = list(levels.values()) if isinstance(levels, dict) else list(levels)
+            return {"type": "score", "score": sum(i * v for i, v in enumerate(p)), "confidence": max(p),
+                    "legend": {str(i): (levels[i] if i < len(levels) else str(i)) for i in range(len(p))},
+                    "probabilities": {str(i): v for i, v in enumerate(p)}}
+        dist = dict(zip(ids, p))
+        best = max(dist, key=dist.get)
+        return {"type": "choice", "choice": best, "confidence": dist[best], "probabilities": dist}
+
+    def _split(self, state, questions):
+        """-> (answers decided without the model, [(key, rendered row)], tokens spent on wide lists)"""
+        answers, rows, tokens = {}, [], 0
+        for k, q in questions.items():
+            t = q.get("type", "choice")
+            crit = q.get("criteria") or {}
+            if t == "choice" and isinstance(crit, dict) and len(crit) == 1:
+                only = next(iter(crit))
+                answers[k] = {"type": "choice", "choice": only, "confidence": 1.0, "probabilities": {only: 1.0}}
+            elif t == "choice" and isinstance(crit, dict) and len(crit) > J.MAX_OPTIONS:
+                p, n = self._wide(state, q)
+                tokens += n
+                best = max(p, key=p.get)
+                answers[k] = {"type": "choice", "choice": best, "confidence": p[best], "probabilities": p}
+            else:
+                rows.append((k, J.from_systemone(state, q)))
+        return answers, rows, tokens
+
+    def decide(self, state, questions):
+        """One request -> (answers, usage), answers in the TypeSafe /v1/systemone format."""
+        answers, rows, tokens = self._split(state, questions)
+        if rows:
+            probs, n = self._probs([r for _, r in rows])
+            tokens += n
+            for (k, row), p in zip(rows, probs):
+                answers[k] = self._answer(row, p, questions[k])
+        return answers, {"input_tokens": tokens, "output_tokens": 0}
+
+    def decide_batch(self, requests):
+        """[(state, questions), ...] -> [answers, ...].  Every question of every request goes through one length-sorted
+        set of padded forward passes (up to max_batch_tokens each), which is much faster than calling decide() in a loop
+        when the requests are short.  Answers are the same as decide() up to bf16 rounding."""
+        parts, flat = [], []
+        for state, questions in requests:
+            answers, rows, _ = self._split(state, questions)
+            parts.append((answers, rows, questions))
+            flat.extend(r for _, r in rows)
+        graphs, self.graphs = self.graphs, {}        # batched eager path; graphs are sized for one request
+        try:
+            probs, _ = self._probs(flat) if flat else ([], 0)
+        finally:
+            self.graphs = graphs
+        out, i = [], 0
+        for answers, rows, questions in parts:
+            for k, row in rows:
+                answers[k] = self._answer(row, probs[i], questions[k])
+                i += 1
+            out.append(answers)
+        return out

startlux_decision/server.py

@@ -0,0 +1,96 @@
+"""TypeSafe-compatible decision server.
+
+    python -m startlux_decision.server --model /path/to/StartLux-Decision-4B --port 8090
+
+POST /v1/systemone  {"state": ..., "questions": {key: {"type", "instructions", "criteria"}}}
+  -> {"answers": {key: answer}, "usage": {"input_tokens", "output_tokens"}, "model": ...}
+GET  /health        {"status", "model", "fast_kernels", "cuda_graphs"}
+GET  /v1/models
+Requests are served one at a time on one GPU.  At start-up one request runs through both the CUDA-graph path and the
+eager path; if their probabilities differ by more than 0.02 the graphs are dropped.
+"""
+import argparse
+import json
+import os
+import threading
+import time
+from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
+
+
+def main():
+    ap = argparse.ArgumentParser()
+    ap.add_argument("--model", required=True, help="local StartLux-Decision directory")
+    ap.add_argument("--host", default="127.0.0.1")
+    ap.add_argument("--port", type=int, default=8090)
+    ap.add_argument("--device", help="cuda or cpu (default: cuda when available)")
+    ap.add_argument("--name", help="model name reported in responses (default: the directory name)")
+    a = ap.parse_args()
+    from .model import StartLuxDecision
+
+    engine = StartLuxDecision(a.model, device=a.device)
+    demo_state = {"ticket": "I was charged twice for order #4411 and the app still shows it as unpaid."}
+    demo_questions = {
+        "team": {"type": "choice", "instructions": "Which team should handle this ticket?",
+                 "criteria": {"billing": "Payments, refunds and invoices", "shipping": "Delivery and tracking",
+                              "technical": "App, login and account problems"}},
+        "urgent": {"type": "noul", "instructions": "Should this ticket be answered today?"},
+        "severity": {"type": "score", "instructions": "How severe is the impact?",
+                     "criteria": ["cosmetic", "annoying", "blocks the customer"]},
+    }
+    engine.decide(demo_state, demo_questions)                     # warm-up
+    if engine.graphs:
+        diff = engine.self_test(demo_state, demo_questions)
+        print(f"graph self-test: max |p_graph - p_eager| = {diff:.2e}", flush=True)
+        if diff > 0.02:
+            print("graph readout disagrees with the eager path, graphs disabled", flush=True)
+            engine.graphs = {}
+    lock = threading.Lock()
+    name = a.name or os.path.basename(os.path.normpath(a.model))
+
+    class Handler(BaseHTTPRequestHandler):
+        def _send(self, code, obj):
+            data = json.dumps(obj).encode()
+            self.send_response(code)
+            self.send_header("Content-Type", "application/json")
+            self.send_header("Content-Length", str(len(data)))
+            self.end_headers()
+            self.wfile.write(data)
+
+        def do_GET(self):
+            path = self.path.rstrip("/")
+            if path in ("/health", "/v1/health"):
+                return self._send(200, {"status": "ok", "model": name, "fast_kernels": engine.fast_kernels,
+                                        "cuda_graphs": len(engine.graphs)})
+            if path == "/v1/models":
+                return self._send(200, {"models": [{"name": name, "description": "StartLux-Decision typed decision model"}]})
+            self._send(404, {"error": "not found"})
+
+        def do_POST(self):
+            if self.path.rstrip("/") != "/v1/systemone":
+                return self._send(404, {"error": "not found"})
+            try:
+                body = json.loads(self.rfile.read(int(self.headers.get("Content-Length") or 0)) or b"{}")
+                questions = body.get("questions")
+                if not isinstance(questions, dict) or not questions:
+                    raise ValueError("questions must be a non-empty object")
+            except ValueError as e:
+                return self._send(400, {"error": str(e)})
+            t = time.perf_counter()
+            try:
+                with lock:
+                    answers, usage = engine.decide(body.get("state"), questions)
+            except ValueError as e:          # unknown question type, prompt over the context limit
+                return self._send(422, {"error": str(e)})
+            self._send(200, {"answers": answers, "usage": usage, "model": name,
+                             "latency_ms": round(1000 * (time.perf_counter() - t), 2)})
+
+        def log_message(self, *args):
+            pass
+
+    print(f"{name} serving on http://{a.host}:{a.port}/v1/systemone (fast kernels: {engine.fast_kernels}, "
+          f"cuda graphs: {len(engine.graphs)})", flush=True)
+    ThreadingHTTPServer((a.host, a.port), Handler).serve_forever()
+
+
+if __name__ == "__main__":
+    main()

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0997f410c57a1f4e53b09e4be8f4a172d90edd9564368fb0847030937229b9f3
+size 12809320

tokenizer_config.json

@@ -0,0 +1,305 @@
+{
+    "add_prefix_space": false,
+    "added_tokens_decoder": {
+        "248044": {
+            "content": "<|endoftext|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248045": {
+            "content": "<|im_start|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248046": {
+            "content": "<|im_end|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248047": {
+            "content": "<|object_ref_start|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248048": {
+            "content": "<|object_ref_end|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248049": {
+            "content": "<|box_start|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248050": {
+            "content": "<|box_end|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248051": {
+            "content": "<|quad_start|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248052": {
+            "content": "<|quad_end|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248053": {
+            "content": "<|vision_start|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248054": {
+            "content": "<|vision_end|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248055": {
+            "content": "<|vision_pad|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248056": {
+            "content": "<|image_pad|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248057": {
+            "content": "<|video_pad|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248058": {
+            "content": "<tool_call>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248059": {
+            "content": "</tool_call>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248060": {
+            "content": "<|fim_prefix|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248061": {
+            "content": "<|fim_middle|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248062": {
+            "content": "<|fim_suffix|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248063": {
+            "content": "<|fim_pad|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248064": {
+            "content": "<|repo_name|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248065": {
+            "content": "<|file_sep|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248066": {
+            "content": "<tool_response>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248067": {
+            "content": "</tool_response>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248068": {
+            "content": "<think>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248069": {
+            "content": "</think>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248070": {
+            "content": "<|audio_start|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248071": {
+            "content": "<|audio_end|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248072": {
+            "content": "<tts_pad>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248073": {
+            "content": "<tts_text_bos>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248074": {
+            "content": "<tts_text_eod>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248075": {
+            "content": "<tts_text_bos_single>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248076": {
+            "content": "<|audio_pad|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        }
+    },
+    "additional_special_tokens": [
+        "<|im_start|>",
+        "<|im_end|>",
+        "<|object_ref_start|>",
+        "<|object_ref_end|>",
+        "<|box_start|>",
+        "<|box_end|>",
+        "<|quad_start|>",
+        "<|quad_end|>",
+        "<|vision_start|>",
+        "<|vision_end|>",
+        "<|vision_pad|>",
+        "<|image_pad|>",
+        "<|video_pad|>"
+    ],
+    "bos_token": null,
+    "chat_template": "{%- set image_count = namespace(value=0) %}\n{%- set video_count = namespace(value=0) %}\n{%- macro render_content(content, do_vision_count, is_system_content=false) %}\n    {%- if content is string %}\n        {{- content }}\n    {%- elif content is iterable and content is not mapping %}\n        {%- for item in content %}\n            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}\n                {%- if is_system_content %}\n                    {{- raise_exception('System message cannot contain images.') }}\n                {%- endif %}\n                {%- if do_vision_count %}\n                    {%- set image_count.value = image_count.value + 1 %}\n                {%- endif %}\n                {%- if add_vision_id %}\n                    {{- 'Picture ' ~ image_count.value ~ ': ' }}\n                {%- endif %}\n                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}\n            {%- elif 'video' in item or item.type == 'video' %}\n                {%- if is_system_content %}\n                    {{- raise_exception('System message cannot contain videos.') }}\n                {%- endif %}\n                {%- if do_vision_count %}\n                    {%- set video_count.value = video_count.value + 1 %}\n                {%- endif %}\n                {%- if add_vision_id %}\n                    {{- 'Video ' ~ video_count.value ~ ': ' }}\n                {%- endif %}\n                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}\n            {%- elif 'text' in item %}\n                {{- item.text }}\n            {%- else %}\n                {{- raise_exception('Unexpected item type in content.') }}\n            {%- endif %}\n        {%- endfor %}\n    {%- elif content is none or content is undefined %}\n        {{- '' }}\n    {%- else %}\n        {{- raise_exception('Unexpected content type.') }}\n    {%- endif %}\n{%- endmacro %}\n{%- if not messages %}\n    {{- raise_exception('No messages provided.') }}\n{%- endif %}\n{%- set reasoning_instructions = '' %}\n{%- if enable_thinking is undefined or enable_thinking is true %}\n    {%- set resolved_reasoning_effort = reasoning_effort|default('xhigh') %}\n    {%- if resolved_reasoning_effort not in ('xhigh', 'medium', 'low') %}\n        {{- raise_exception('Unexpected reasoning effort ' ~ reasoning_effort ~ '. Supported types are xhigh (default), medium, and low.') }}\n    {%- endif %}\n    {%- if resolved_reasoning_effort == 'xhigh' %}\n        {%- set reasoning_instructions = 'Reasoning effort is set to xhigh. Please think carefully through the task, validate key assumptions, consider plausible alternatives, and prioritize correctness, consistency, and clarity in the final answer.' %}\n    {%- elif resolved_reasoning_effort == 'low' %}\n        {%- set reasoning_instructions = 'Reasoning effort is set to low. Keep your thinking brief and focused, moving directly to the conclusion without unnecessary elaboration.' %}\n    {%- endif %}\n{%- endif %}\n{%- if tools and tools is iterable and tools is not mapping %}\n    {{- '<|im_start|>system\\n' }}\n    {%- if reasoning_instructions %}\n        {{- reasoning_instructions + '\\n\\n' }}\n    {%- endif %}\n    {{- \"# Tools\\n\\nYou have access to the following functions:\\n\\n<tools>\" }}\n    {%- for tool in tools %}\n        {{- \"\\n\" }}\n        {{- tool | tojson }}\n    {%- endfor %}\n    {{- \"\\n</tools>\" }}\n    {{- '\\n\\nIf you choose to call a function ONLY reply in the following format with NO suffix:\\n\\n<tool_call>\\n<function=example_function_name>\\n<parameter=example_parameter_1>\\nvalue_1\\n</parameter>\\n<parameter=example_parameter_2>\\nThis is the value for the second parameter\\nthat can span\\nmultiple lines\\n</parameter>\\n</function>\\n</tool_call>\\n\\n<IMPORTANT>\\nReminder:\\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\\n- Required parameters MUST be specified\\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\\n</IMPORTANT>' }}\n    {%- if messages[0].role == 'system' %}\n        {%- set content = render_content(messages[0].content, false, true)|trim %}\n        {%- if content %}\n            {{- '\\n\\n' + content }}\n        {%- endif %}\n    {%- endif %}\n    {{- '<|im_end|>\\n' }}\n{%- else %}\n    {%- if messages[0].role == 'system' %}\n        {%- set content = render_content(messages[0].content, false, true)|trim %}\n        {%- if content %}\n            {{- '<|im_start|>system\\n' + (reasoning_instructions + '\\n\\n' if reasoning_instructions else '')  + content + '<|im_end|>\\n' }}\n        {%- elif reasoning_instructions %}\n            {{- '<|im_start|>system\\n' + reasoning_instructions + '<|im_end|>\\n' }}\n        {%- endif %}\n    {%- elif reasoning_instructions %}\n        {{- '<|im_start|>system\\n' + reasoning_instructions + '<|im_end|>\\n' }}\n    {%- endif %}\n{%- endif %}\n{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}\n{%- for message in messages[::-1] %}\n    {%- set index = (messages|length - 1) - loop.index0 %}\n    {%- if ns.multi_step_tool and message.role == \"user\" %}\n        {%- set content = render_content(message.content, false)|trim %}\n        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}\n            {%- set ns.multi_step_tool = false %}\n            {%- set ns.last_query_index = index %}\n        {%- endif %}\n    {%- endif %}\n{%- endfor %}\n{%- if ns.multi_step_tool %}\n    {{- raise_exception('No user query found in messages.') }}\n{%- endif %}\n{%- for message in messages %}\n    {%- set content = render_content(message.content, true)|trim %}\n    {%- if message.role == \"system\" %}\n        {%- if not loop.first %}\n            {{- raise_exception('System message must be at the beginning.') }}\n        {%- endif %}\n    {%- elif message.role == \"user\" %}\n        {{- '<|im_start|>' + message.role + '\\n' + content + '<|im_end|>' + '\\n' }}\n    {%- elif message.role == \"assistant\" %}\n        {%- set reasoning_content = '' %}\n        {%- if message.reasoning_content is string %}\n            {%- set reasoning_content = message.reasoning_content %}\n        {%- endif %}\n        {%- set reasoning_content = reasoning_content|trim %}\n        {%- if preserve_thinking is undefined or preserve_thinking is true or loop.index0 > ns.last_query_index %}\n            {{- '<|im_start|>' + message.role + '\\n<think>\\n' + reasoning_content + '\\n</think>\\n\\n' + content }}\n        {%- else %}\n            {{- '<|im_start|>' + message.role + '\\n' + content }}\n        {%- endif %}\n        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}\n            {%- for tool_call in message.tool_calls %}\n                {%- if tool_call.function is defined %}\n                    {%- set tool_call = tool_call.function %}\n                {%- endif %}\n                {%- if loop.first %}\n                    {%- if content|trim %}\n                        {{- '\\n\\n<tool_call>\\n<function=' + tool_call.name + '>\\n' }}\n                    {%- else %}\n                        {{- '<tool_call>\\n<function=' + tool_call.name + '>\\n' }}\n                    {%- endif %}\n                {%- else %}\n                    {{- '\\n<tool_call>\\n<function=' + tool_call.name + '>\\n' }}\n                {%- endif %}\n                {%- if tool_call.arguments is defined and tool_call.arguments != '' %}\n                    {%- for args_name, args_value in tool_call.arguments|items %}\n                        {{- '<parameter=' + args_name + '>\\n' }}\n                        {%- set args_value = args_value | string if args_value is string else args_value | tojson | safe %}\n                        {{- args_value }}\n                        {{- '\\n</parameter>\\n' }}\n                    {%- endfor %}\n                {%- endif %}\n                {{- '</function>\\n</tool_call>' }}\n            {%- endfor %}\n        {%- endif %}\n        {{- '<|im_end|>\\n' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if loop.previtem and loop.previtem.role != \"tool\" %}\n            {{- '<|im_start|>user' }}\n        {%- endif %}\n        {{- '\\n<tool_response>\\n' }}\n        {{- content }}\n        {{- '\\n</tool_response>' }}\n        {%- if not loop.last and loop.nextitem.role != \"tool\" %}\n            {{- '<|im_end|>\\n' }}\n        {%- elif loop.last %}\n            {{- '<|im_end|>\\n' }}\n        {%- endif %}\n    {%- else %}\n        {{- raise_exception('Unexpected message role.') }}\n    {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n    {{- '<|im_start|>assistant\\n' }}\n    {%- if enable_thinking is defined and enable_thinking is false %}\n        {{- '<think>\\n\\n</think>\\n\\n' }}\n    {%- else %}\n        {{- '<think>\\n' }}\n    {%- endif %}\n{%- endif %}",
+    "clean_up_tokenization_spaces": false,
+    "eos_token": "<|im_end|>",
+    "errors": "replace",
+    "model_max_length": 262144,
+    "pad_token": "<|endoftext|>",
+    "split_special_tokens": false,
+    "tokenizer_class": "Qwen2Tokenizer",
+    "unk_token": null,
+    "add_bos_token": false,
+    "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+    "extra_special_tokens": {
+        "audio_bos_token": "<|audio_start|>",
+        "audio_eos_token": "<|audio_end|>",
+        "audio_token": "<|audio_pad|>",
+        "image_token": "<|image_pad|>",
+        "video_token": "<|video_pad|>",
+        "vision_bos_token": "<|vision_start|>",
+        "vision_eos_token": "<|vision_end|>"
+    }
+}