Publish verified RWKV7 G1k 2.9B release

LICENSE

@@ -0,0 +1,201 @@
+                                 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.

NOTICE

@@ -0,0 +1,8 @@
+RWKV-7 G1k model release
+
+Source checkpoint: BlinkDL/rwkv7-g1/rwkv7-g1k-2.9b-20260930-ctx25600.pth
+Source revision: a1ddf9e96df23d5d7db65137ad83a4701044cd05
+Source SHA-256: d8f8ecd4af8d706f867ac8a74284f35c0e3b9f863e7b09bb4d4031480c8b28b2
+FLA implementation revision: 8e84ed4a6727be082c34a3855c60623fd11411e9
+
+The RWKV logo in assets/rwkv-logo.webp is the audited Hugging Face RWKV organization avatar snapshot. RWKV names and logos may be subject to separate trademark rules; Apache-2.0 does not grant trademark rights.

README.md

@@ -0,0 +1,153 @@
+---
+license: apache-2.0
+library_name: transformers
+pipeline_tag: text-generation
+base_model: BlinkDL/rwkv7-g1
+language:
+- en
+- zh
+- fr
+- es
+- de
+- pt
+- ru
+- it
+- ja
+- ko
+- vi
+- ar
+datasets:
+- HuggingFaceFW/fineweb-edu
+- mlfoundations/dclm-baseline-1.0
+- cerebras/SlimPajama-627B
+- EleutherAI/pile
+- bigcode/starcoderdata
+- oscar-corpus/OSCAR-2301
+tags:
+- rwkv
+- rwkv7
+- fla
+- recurrent
+- causal-lm
+---
+
+<!-- markdownlint-disable first-line-h1 -->
+<!-- markdownlint-disable html -->
+
+<div align="center">
+  <a href="https://www.rwkv.com/">
+    <img src="assets/rwkv-logo.webp" width="140" alt="RWKV logo" />
+  </a>
+  <h1>RWKV7-G1k-2.9B-20260930</h1>
+  <p><strong>RWKV-7 “Goose” in Flash Linear Attention format</strong></p>
+</div>
+
+This repository provides the **RWKV-7 G1k 2.9B** checkpoint in the
+[`flash-linear-attention`](https://github.com/fla-org/flash-linear-attention)
+(FLA) RWKV7 layout for Transformers-compatible inference.
+
+> [!IMPORTANT]
+> This is a base language model, not a safety-aligned instruction-tuned
+> assistant. The included chat template provides a conversational prompt
+> format, but the model may not follow instructions consistently.
+
+## About RWKV-7
+
+RWKV-7, also called **Goose**, is an attention-free recurrent language model.
+It maintains a constant-size recurrent state instead of an attention KV cache
+that grows with the preceding sequence. Training remains parallelizable, while
+recurrent decoding uses constant state size and constant work per generated
+token with respect to sequence length.
+
+`G1k` identifies this checkpoint revision. The source checkpoint is available
+from [`BlinkDL/rwkv7-g1`](https://huggingface.co/BlinkDL/rwkv7-g1).
+
+## Model details
+
+| Property | Value |
+| --- | --- |
+| Architecture | RWKV-7 G1k |
+| Parameters in this FLA checkpoint | 2.948B |
+| Layers | 32 |
+| Hidden size | 2,560 |
+| Heads | 40 × 64 |
+| Feed-forward size | 10,240 |
+| Vocabulary | RWKV World tokenizer, 65,536 tokens |
+| Configured context length | 25,600 tokens |
+| Weight dtype | BF16 |
+| License | Apache-2.0 |
+
+## Run with FLA
+
+Use a CUDA-capable NVIDIA GPU with BF16 support. Install the audited FLA
+revision with its CUDA dependency extra:
+
+```bash
+python -m pip install \
+  "flash-linear-attention[cuda] @ git+https://github.com/fla-org/flash-linear-attention.git@8e84ed4a6727be082c34a3855c60623fd11411e9" \
+  "transformers>=4.50.2"
+```
+
+Import `fla` before using the Auto classes so that the RWKV7 implementation is
+registered:
+
+```python
+import fla
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer, PretrainedConfig
+
+model_id = "shoumenchougou/RWKV7-G1k-2.9B-20260930"
+
+tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True, config=PretrainedConfig())
+model = AutoModelForCausalLM.from_pretrained(
+    model_id,
+    torch_dtype=torch.bfloat16,
+).to("cuda").eval()
+
+messages = [
+    {"role": "user", "content": "Explain recurrent language models in one paragraph."}
+]
+prompt = tokenizer.apply_chat_template(
+    messages,
+    tokenize=False,
+    add_generation_prompt=True,
+    thinking=False,
+)
+inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
+
+with torch.inference_mode():
+    output_ids = model.generate(
+        **inputs,
+        max_new_tokens=128,
+        do_sample=True,
+        temperature=0.7,
+        top_p=0.9,
+    )
+
+new_tokens = output_ids[0, inputs.input_ids.shape[1]:]
+print(tokenizer.decode(new_tokens, skip_special_tokens=True))
+```
+
+Set `thinking=True` when applying the chat template to leave an open thinking
+prefix. This changes only the prompt format; it does not turn the base model
+into an instruction-tuned or safety-aligned assistant.
+
+## Compatibility and validation
+
+The package was checked with FLA commit
+`8e84ed4a6727be082c34a3855c60623fd11411e9` and Transformers 4.50.2. Its
+configuration, tokenizer, chat template, BF16 weights, and Transformers model
+loading were validated locally. All 1,059 model tensors load into
+`RWKV7ForCausalLM` without missing, unexpected, or mismatched keys.
+
+CUDA/Triton generation was not executed in the local CPU-only validation
+environment. Runtime behavior can depend on the GPU, CUDA, PyTorch, Triton,
+and FLA versions. Cross-check benchmark or evaluation results against the
+official RWKV implementation before reporting them.
+
+## References
+
+- [RWKV-7 paper](https://arxiv.org/abs/2503.14456)
+- [RWKV-LM](https://github.com/BlinkDL/RWKV-LM)
+- [Flash Linear Attention](https://github.com/fla-org/flash-linear-attention)
+- [Source checkpoint repository](https://huggingface.co/BlinkDL/rwkv7-g1)

chat_template.jinja

@@ -0,0 +1,58 @@
+{%- set add_generation_prompt = add_generation_prompt | default(false) -%}
+{%- set thinking = thinking | default(false) -%}
+{%- set bos_token = bos_token | default('', true) -%}
+{%- set tools = tools | default([], true) -%}
+{%- set ns = namespace(system_prompt='') -%}
+{%- for message in messages -%}
+    {%- if message.role == 'system' -%}
+        {%- set ns.system_prompt = message.content | trim -%}
+    {%- endif -%}
+{%- endfor -%}
+{{- bos_token -}}
+{%- if ns.system_prompt or tools | length > 0 -%}
+{{ 'System: ' }}{{ ns.system_prompt }}
+{%- if tools | length > 0 -%}
+{%- if ns.system_prompt %}{{ '\n' }}{%- endif -%}
+{{ 'Tools:\n' -}}
+{{ tools | tojson }}
+{{ '\nWhen using a tool, return only a compact JSON function call in a ```json block, like {"name":"calculator","arguments":{"expression":"2+2"}}. The `name` field must be top-level, never inside `arguments`. Do not copy the tool schema into arguments. Otherwise answer normally.' }}
+{%- endif -%}
+{{ '\n\n' }}
+{%- endif -%}
+{%- for message in messages -%}
+{%- if message.role == 'user' -%}
+{{ 'User: ' ~ (message.content | trim) ~ '\n\n' }}
+{%- elif message.role == 'assistant' -%}
+{%- generation -%}
+{%- set content = message.content | default('', true) | trim -%}
+{{ 'Assistant:' }}
+{%- if message.tool_calls is defined and message.tool_calls | length > 0 -%}
+{%- if content %}{{ ' ' ~ content ~ '\n' }}{%- endif -%}
+{%- for tool_call in message.tool_calls -%}
+    {%- if tool_call.function is defined -%}
+        {%- set name = tool_call.function.name | default('') -%}
+        {%- set args = tool_call.function.arguments | default({}, true) -%}
+    {%- else -%}
+        {%- set name = tool_call.name | default('') -%}
+        {%- set args = tool_call.arguments | default({}, true) -%}
+    {%- endif -%}
+{{ ' ```json\n' -}}
+{{ '{"name": ' }}{{ name | tojson }}{{ ', "arguments": ' }}{% if args is string %}{{ args }}{% else %}{{ args | tojson }}{% endif %}{{ '}\n' -}}
+{{ '```' }}{{ '\n' if not loop.last else '' }}
+{%- endfor -%}
+{%- elif content -%}
+{{ ' ' ~ content }}
+{%- endif -%}
+{{ '\n\n' }}
+{%- endgeneration -%}
+{%- elif message.role == 'tool' -%}
+{{ 'User: Function output:\n' ~ (message.content | trim) ~ '\n\n' }}
+{%- endif -%}
+{%- endfor -%}
+{%- if add_generation_prompt -%}
+{%- if thinking -%}
+{{ 'Assistant: <think' }}
+{%- else -%}
+{{ 'Assistant: <think></think>\n' }}
+{%- endif -%}
+{%- endif -%}

config.json

@@ -0,0 +1,70 @@
+{
+  "a_low_rank_dim": 96,
+  "architectures": [
+    "RWKV7ForCausalLM"
+  ],
+  "attn": null,
+  "attn_mode": "chunk",
+  "bos_token_id": 0,
+  "decay_low_rank_dim": 96,
+  "eos_token_id": 0,
+  "fuse_cross_entropy": true,
+  "fuse_linear_cross_entropy": false,
+  "fuse_norm": true,
+  "gate_low_rank_dim": 320,
+  "head_dim": 64,
+  "hidden_act": "sqrelu",
+  "hidden_ratio": 4.0,
+  "hidden_size": 2560,
+  "initializer_range": 0.02,
+  "intermediate_size": 10240,
+  "max_position_embeddings": 25600,
+  "model_type": "rwkv7",
+  "norm_bias": true,
+  "norm_eps": 1e-05,
+  "norm_first": true,
+  "num_heads": 40,
+  "num_hidden_layers": 32,
+  "tie_word_embeddings": false,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.50.2",
+  "use_cache": true,
+  "use_l2warp": true,
+  "v_low_rank_dim": 64,
+  "value_dim": [
+    2560,
+    2560,
+    2560,
+    2560,
+    2560,
+    2560,
+    2560,
+    2560,
+    2560,
+    2560,
+    2560,
+    2560,
+    2560,
+    2560,
+    2560,
+    2560,
+    2560,
+    2560,
+    2560,
+    2560,
+    2560,
+    2560,
+    2560,
+    2560,
+    2560,
+    2560,
+    2560,
+    2560,
+    2560,
+    2560,
+    2560,
+    2560
+  ],
+  "vocab_size": 65536,
+  "pad_token_id": 0
+}

generation_config.json

@@ -0,0 +1,6 @@
+{
+  "bos_token_id": 0,
+  "eos_token_id": 0,
+  "pad_token_id": 0,
+  "use_cache": true
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:92b14263e2c78aa2a07b636e23395726f15ce3b53e2a5cfaea6681ec7215f4fb
+size 5895584920

tokenization_rwkv7.py

@@ -0,0 +1,343 @@
+# coding=utf-8
+"""Exact linear-time RWKV World tokenizer for Transformers AutoTokenizer."""
+
+from __future__ import annotations
+
+import json
+import os
+import shutil
+from typing import Optional
+
+from transformers.tokenization_utils import PreTrainedTokenizer
+
+
+VOCAB_FILES_NAMES = {"tokenizer_file": "tokenizer.json"}
+END_TOKEN = "<|endoftext|>"
+INVALID_TOKEN_PREFIX = "\ue000"
+INVALID_TOKEN_SUFFIX = "\ue001"
+
+
+class _CharEncoding:
+    """Minimal Encoding surface used by BatchEncoding.char_to_token."""
+
+    def __init__(self, offsets):
+        self.offsets = offsets
+
+    def char_to_token(self, char_index, sequence_index=0):
+        if sequence_index != 0:
+            return None
+        for token_index, offset in enumerate(self.offsets):
+            if offset is not None and offset[0] <= char_index < offset[1]:
+                return token_index
+        return None
+
+
+def _bytes_to_unicode():
+    values = list(range(ord("!"), ord("~") + 1))
+    values += list(range(ord("¡"), ord("¬") + 1))
+    values += list(range(ord("®"), ord("ÿ") + 1))
+    characters = list(values)
+    extra = 0
+    for byte in range(256):
+        if byte not in values:
+            values.append(byte)
+            characters.append(256 + extra)
+            extra += 1
+    return dict(zip(values, map(chr, characters)))
+
+
+class Rwkv7Tokenizer(PreTrainedTokenizer):
+    """RWKV World longest-prefix tokenizer backed only by tokenizer.json."""
+
+    vocab_files_names = VOCAB_FILES_NAMES
+    model_input_names = ["input_ids", "attention_mask"]
+    padding_side = "left"
+
+    def __init__(
+        self,
+        tokenizer_file,
+        eos_token=END_TOKEN,
+        pad_token=END_TOKEN,
+        unk_token=END_TOKEN,
+        add_bos_token=False,
+        **kwargs,
+    ):
+        if not tokenizer_file or not os.path.isfile(tokenizer_file):
+            raise ValueError(f"tokenizer.json does not exist: {tokenizer_file}")
+        if add_bos_token:
+            raise ValueError("RWKV World has no BOS token")
+        self.tokenizer_file = tokenizer_file
+        self.add_bos_token = False
+        value = json.loads(open(tokenizer_file, encoding="utf-8").read())
+        model = value.get("model")
+        if not isinstance(model, dict) or model.get("type") != "WordPiece":
+            raise ValueError("tokenizer.json is not the locked RWKV WordPiece model")
+        if (
+            model.get("unk_token") != END_TOKEN
+            or model.get("continuing_subword_prefix") != ""
+            or value.get("normalizer") is not None
+        ):
+            raise ValueError("tokenizer.json does not use exact RWKV semantics")
+        pre_tokenizer = value.get("pre_tokenizer")
+        decoder = value.get("decoder")
+        if (
+            not isinstance(pre_tokenizer, dict)
+            or pre_tokenizer.get("type") != "ByteLevel"
+            or pre_tokenizer.get("add_prefix_space") is not False
+            or pre_tokenizer.get("use_regex") is not False
+            or not isinstance(decoder, dict)
+            or decoder.get("type") != "ByteLevel"
+        ):
+            raise ValueError("tokenizer.json does not use exact ByteLevel semantics")
+        vocabulary = model.get("vocab")
+        if not isinstance(vocabulary, dict) or not vocabulary:
+            raise ValueError("tokenizer.json has no vocabulary")
+        if any(not isinstance(token, str) or not isinstance(index, int) for token, index in vocabulary.items()):
+            raise TypeError("tokenizer.json vocabulary entry is invalid")
+        if set(vocabulary.values()) != set(range(len(vocabulary))):
+            raise ValueError("tokenizer.json vocabulary ids are not contiguous")
+        if vocabulary.get(END_TOKEN) != 0:
+            raise ValueError("tokenizer.json must reserve id 0 for end-of-text")
+
+        byte_decoder = {character: byte for byte, character in _bytes_to_unicode().items()}
+        self.encoder = dict(vocabulary)
+        self.decoder = {index: token for token, index in vocabulary.items()}
+        self._token_bytes = {}
+        reverse = {}
+        for token, index in vocabulary.items():
+            if index == 0:
+                continue
+            placeholder = f"{INVALID_TOKEN_PREFIX}{index}{INVALID_TOKEN_SUFFIX}"
+            if token == placeholder:
+                continue
+            try:
+                raw = bytes(byte_decoder[character] for character in token)
+            except KeyError as error:
+                raise ValueError("tokenizer.json vocabulary is not ByteLevel encoded") from error
+            if raw in reverse:
+                raise ValueError("tokenizer.json contains duplicate byte tokens")
+            reverse[raw] = index
+            self._token_bytes[token] = raw
+        missing = [byte for byte in range(256) if bytes([byte]) not in reverse]
+        if missing:
+            raise ValueError(f"tokenizer.json is missing singleton bytes: {missing}")
+        for raw, index in reverse.items():
+            for endpoint in range(1, len(raw)):
+                prefix = reverse.get(raw[:endpoint])
+                if prefix is not None and prefix > index:
+                    raise ValueError("token ranks do not support longest-prefix encoding")
+
+        self._children = [{}]
+        self._terminal = [None]
+        self._max_token_bytes = 1
+        for raw, index in sorted(reverse.items(), key=lambda item: item[1]):
+            node = 0
+            for byte in raw:
+                child = self._children[node].get(byte)
+                if child is None:
+                    child = len(self._children)
+                    self._children[node][byte] = child
+                    self._children.append({})
+                    self._terminal.append(None)
+                node = child
+            self._terminal[node] = index
+            self._max_token_bytes = max(self._max_token_bytes, len(raw))
+
+        if "additional_special_tokens" not in kwargs:
+            appended = []
+            for item in sorted(value.get("added_tokens", []), key=lambda item: item.get("id", -1)):
+                if not isinstance(item, dict):
+                    raise TypeError("tokenizer.json added token is invalid")
+                content = item.get("content")
+                index = item.get("id")
+                if not isinstance(content, str) or not isinstance(index, int):
+                    raise TypeError("tokenizer.json added token is invalid")
+                if content != END_TOKEN:
+                    if not item.get("special") or index < len(vocabulary):
+                        raise ValueError("only append-only special tokens are supported")
+                    appended.append(content)
+            if appended:
+                kwargs["additional_special_tokens"] = appended
+        super().__init__(
+            eos_token=eos_token,
+            pad_token=pad_token,
+            unk_token=unk_token,
+            add_bos_token=self.add_bos_token,
+            **kwargs,
+        )
+
+    @property
+    def vocab_size(self):
+        return len(self.encoder)
+
+    def get_vocab(self):
+        vocabulary = dict(self.encoder)
+        vocabulary.update(self.added_tokens_encoder)
+        return vocabulary
+
+    def _encode_bytes(self, data):
+        output = []
+        position = 0
+        while position < len(data):
+            node = 0
+            cursor = position
+            best_id = None
+            best_end = position
+            limit = min(len(data), position + self._max_token_bytes)
+            while cursor < limit:
+                child = self._children[node].get(data[cursor])
+                if child is None:
+                    break
+                node = child
+                cursor += 1
+                token_id = self._terminal[node]
+                if token_id is not None and (best_id is None or token_id > best_id):
+                    best_id = token_id
+                    best_end = cursor
+            if best_id is None:
+                raise RuntimeError("singleton-byte vocabulary invariant was violated")
+            output.append(best_id)
+            position = best_end
+        return output
+
+    def _encode_bytes_with_offsets(self, data, char_offset, byte_to_char):
+        output = []
+        position = 0
+        while position < len(data):
+            node = 0
+            cursor = position
+            best_id = None
+            best_end = position
+            limit = min(len(data), position + self._max_token_bytes)
+            while cursor < limit:
+                child = self._children[node].get(data[cursor])
+                if child is None:
+                    break
+                node = child
+                cursor += 1
+                token_id = self._terminal[node]
+                if token_id is not None and (best_id is None or token_id > best_id):
+                    best_id = token_id
+                    best_end = cursor
+            if best_id is None:
+                raise RuntimeError("singleton-byte vocabulary invariant was violated")
+            start_char = char_offset + byte_to_char[position]
+            end_char = char_offset + byte_to_char[best_end - 1] + 1
+            output.append((best_id, (start_char, end_char)))
+            position = best_end
+        return output
+
+    def _encode_with_offsets(self, text):
+        special = sorted(self.added_tokens_encoder, key=lambda token: (-len(token), token))
+        output = []
+        ordinary_start = 0
+        position = 0
+
+        def append_ordinary(segment, char_offset):
+            if not segment:
+                return
+            data = segment.encode("utf-8")
+            byte_to_char = []
+            for char_index, character in enumerate(segment):
+                byte_to_char.extend([char_index] * len(character.encode("utf-8")))
+            output.extend(
+                self._encode_bytes_with_offsets(data, char_offset, byte_to_char)
+            )
+
+        while position < len(text):
+            matched = next(
+                (token for token in special if text.startswith(token, position)), None
+            )
+            if matched is None:
+                position += 1
+                continue
+            append_ordinary(text[ordinary_start:position], ordinary_start)
+            output.append(
+                (
+                    self.added_tokens_encoder[matched],
+                    (position, position + len(matched)),
+                )
+            )
+            position += len(matched)
+            ordinary_start = position
+        append_ordinary(text[ordinary_start:], ordinary_start)
+        return output
+
+    def _tokenize(self, text, **kwargs):
+        del kwargs
+        return [self.decoder[index] for index in self._encode_bytes(text.encode("utf-8"))]
+
+    def __call__(self, text=None, text_pair=None, **kwargs):
+        output = super().__call__(text=text, text_pair=text_pair, **kwargs)
+        if text_pair is not None or text is None:
+            return output
+        texts = [text] if isinstance(text, str) else list(text)
+        if any(not isinstance(item, str) for item in texts):
+            return output
+        input_ids = output["input_ids"]
+        attention_mask = output.get("attention_mask")
+        raw_ids = input_ids.tolist() if hasattr(input_ids, "tolist") else input_ids
+        if raw_ids and isinstance(raw_ids[0], int):
+            rows = [raw_ids]
+            masks = [attention_mask] if attention_mask is not None else None
+        else:
+            rows = raw_ids
+            if attention_mask is None:
+                masks = None
+            else:
+                masks = (
+                    attention_mask.tolist()
+                    if hasattr(attention_mask, "tolist")
+                    else attention_mask
+                )
+        encodings = []
+        for index, source in enumerate(texts):
+            exact = self._encode_with_offsets(source)
+            row = rows[index]
+            active = len(row) if masks is None else sum(int(item) for item in masks[index])
+            exact = exact[:active]
+            offsets = [offset for _token_id, offset in exact]
+            missing = len(row) - len(offsets)
+            if self.padding_side == "left":
+                offsets = [None] * missing + offsets
+            else:
+                offsets.extend([None] * missing)
+            encodings.append(_CharEncoding(offsets))
+        output._encodings = encodings
+        return output
+
+    def _convert_token_to_id(self, token):
+        return self.encoder.get(token, 0)
+
+    def _convert_id_to_token(self, index):
+        return self.decoder.get(index, END_TOKEN)
+
+    def convert_tokens_to_string(self, tokens):
+        decoded = bytearray()
+        for token in tokens:
+            raw = self._token_bytes.get(token)
+            if raw is not None:
+                decoded.extend(raw)
+            else:
+                decoded.extend(str(token).encode("utf-8"))
+        return bytes(decoded).decode("utf-8", errors="replace")
+
+    def build_inputs_with_special_tokens(self, token_ids_0, token_ids_1=None):
+        bos = [self.bos_token_id] if self.add_bos_token else []
+        output = bos + list(token_ids_0)
+        if token_ids_1 is not None:
+            output.extend(bos + list(token_ids_1))
+        return output
+
+    def save_vocabulary(self, save_directory, filename_prefix: Optional[str] = None):
+        if not os.path.isdir(save_directory):
+            raise ValueError("save directory does not exist")
+        filename = (f"{filename_prefix}-" if filename_prefix else "") + "tokenizer.json"
+        destination = os.path.join(save_directory, filename)
+        if os.path.abspath(destination) != os.path.abspath(self.tokenizer_file):
+            shutil.copyfile(self.tokenizer_file, destination)
+        return (destination,)
+
+
+__all__ = ["Rwkv7Tokenizer"]
+

tokenizer_config.json

@@ -0,0 +1,24 @@
+{
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "auto_map": {
+    "AutoTokenizer": [
+      "tokenization_rwkv7.Rwkv7Tokenizer",
+      null
+    ]
+  },
+  "backend": "rwkv_world_trie",
+  "eos_token": "<|endoftext|>",
+  "pad_token": "<|endoftext|>",
+  "padding_side": "left",
+  "tokenizer_class": "Rwkv7Tokenizer",
+  "unk_token": "<|endoftext|>"
+}