Add H200-distilled DFlash drafter GGUFs

bundles/0_8b/dflash/distill-manifest.json

@@ -0,0 +1,129 @@
+{
+  "schemaVersion": 1,
+  "kind": "dflash-drafter-h200-distillation",
+  "tier": "0_8b",
+  "drafterSizeB": 0.1,
+  "vocabSize": 248077,
+  "targetTokenizerSha256": "5905ac13521957e9c864df215308ee2c62fd6d512758768d45e53dcf16bc1ab1",
+  "studentTokenizerSha256": "5905ac13521957e9c864df215308ee2c62fd6d512758768d45e53dcf16bc1ab1",
+  "tokenizerParity": {
+    "target": {
+      "class": "Qwen2Tokenizer",
+      "vocabSize": 248077,
+      "vocabSha256": "467543819daf0b32f85b6b8e9097aa1ac1675affa59836c26acbc4a22eb83f0f",
+      "addedVocabSha256": "5d4b1146a357eb5fc099cce618f7589a4bdf82b6e863731b559e4f6a8726bad6",
+      "specialTokensMap": {
+        "eos_token": "<|endoftext|>",
+        "pad_token": "<|endoftext|>",
+        "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|>"
+      },
+      "allSpecialIds": [
+        248044,
+        248070,
+        248071,
+        248076,
+        248056,
+        248057,
+        248053,
+        248054
+      ],
+      "allSpecialTokens": [
+        "<|endoftext|>",
+        "<|audio_start|>",
+        "<|audio_end|>",
+        "<|audio_pad|>",
+        "<|image_pad|>",
+        "<|video_pad|>",
+        "<|vision_start|>",
+        "<|vision_end|>"
+      ],
+      "chatTemplate": "{%- 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{%- if tools and tools is iterable and tools is not mapping %}\n    {{- '<|im_start|>system\\n' }}\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        {{- '<|im_start|>system\\n' + content + '<|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        {%- else %}\n            {%- if '</think>' in content %}\n                {%- set reasoning_content = content.split('</think>')[0].rstrip('\\n').split('<think>')[-1].lstrip('\\n') %}\n                {%- set content = content.split('</think>')[-1].lstrip('\\n') %}\n            {%- endif %}\n        {%- endif %}\n        {%- set reasoning_content = reasoning_content|trim %}\n        {%- if 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 %}\n                    {%- for args_name, args_value in tool_call.arguments|items %}\n                        {{- '<parameter=' + args_name + '>\\n' }}\n                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}\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 true %}\n        {{- '<think>\\n' }}\n    {%- else %}\n        {{- '<think>\\n\\n</think>\\n\\n' }}\n    {%- endif %}\n{%- endif %}",
+      "files": {
+        "chat_template.jinja": "273d8e0e683b885071fb17e08d71e5f2a5ddfb5309756181681de4f5a1822d80",
+        "tokenizer.json": "06b9509352d2af50381ab2247e083b80d32d5c0aba91c272ca9ff729b6a0e523",
+        "tokenizer_config.json": "8671bed7c852ce9e661be94f179a7b4ffd091c2a65aea0363e5501c20318ee45"
+      },
+      "sha256": "5905ac13521957e9c864df215308ee2c62fd6d512758768d45e53dcf16bc1ab1"
+    },
+    "student": {
+      "class": "Qwen2Tokenizer",
+      "vocabSize": 248077,
+      "vocabSha256": "467543819daf0b32f85b6b8e9097aa1ac1675affa59836c26acbc4a22eb83f0f",
+      "addedVocabSha256": "5d4b1146a357eb5fc099cce618f7589a4bdf82b6e863731b559e4f6a8726bad6",
+      "specialTokensMap": {
+        "eos_token": "<|endoftext|>",
+        "pad_token": "<|endoftext|>",
+        "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|>"
+      },
+      "allSpecialIds": [
+        248044,
+        248070,
+        248071,
+        248076,
+        248056,
+        248057,
+        248053,
+        248054
+      ],
+      "allSpecialTokens": [
+        "<|endoftext|>",
+        "<|audio_start|>",
+        "<|audio_end|>",
+        "<|audio_pad|>",
+        "<|image_pad|>",
+        "<|video_pad|>",
+        "<|vision_start|>",
+        "<|vision_end|>"
+      ],
+      "chatTemplate": "{%- 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{%- if tools and tools is iterable and tools is not mapping %}\n    {{- '<|im_start|>system\\n' }}\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        {{- '<|im_start|>system\\n' + content + '<|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        {%- else %}\n            {%- if '</think>' in content %}\n                {%- set reasoning_content = content.split('</think>')[0].rstrip('\\n').split('<think>')[-1].lstrip('\\n') %}\n                {%- set content = content.split('</think>')[-1].lstrip('\\n') %}\n            {%- endif %}\n        {%- endif %}\n        {%- set reasoning_content = reasoning_content|trim %}\n        {%- if 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 %}\n                    {%- for args_name, args_value in tool_call.arguments|items %}\n                        {{- '<parameter=' + args_name + '>\\n' }}\n                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}\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 true %}\n        {{- '<think>\\n' }}\n    {%- else %}\n        {{- '<think>\\n\\n</think>\\n\\n' }}\n    {%- endif %}\n{%- endif %}",
+      "files": {
+        "chat_template.jinja": "273d8e0e683b885071fb17e08d71e5f2a5ddfb5309756181681de4f5a1822d80",
+        "tokenizer.json": "06b9509352d2af50381ab2247e083b80d32d5c0aba91c272ca9ff729b6a0e523",
+        "tokenizer_config.json": "8671bed7c852ce9e661be94f179a7b4ffd091c2a65aea0363e5501c20318ee45"
+      },
+      "sha256": "5905ac13521957e9c864df215308ee2c62fd6d512758768d45e53dcf16bc1ab1"
+    },
+    "matches": true,
+    "probeEncodingsMatch": true,
+    "targetProbeEncodingsSha256": "03d8f0b7283f474c2748ce5bc19705ca967befb271ed2736f12ed121793ac8b1",
+    "studentProbeEncodingsSha256": "03d8f0b7283f474c2748ce5bc19705ca967befb271ed2736f12ed121793ac8b1"
+  },
+  "studentBase": null,
+  "studentConfig": "/home/ubuntu/eliza-workspace/milady/eliza/packages/training/configs/dflash-drafter-0_1b-qwen3_5",
+  "targetCheckpoint": "Qwen/Qwen3.5-0.8B-Base",
+  "targetGguf": null,
+  "generatedAt": "2026-05-17T08:50:24.055654+00:00",
+  "elapsedSeconds": 788.05628,
+  "synthetic": false,
+  "hyperparameters": {
+    "epochs": 4,
+    "batchSize": 2,
+    "gradAccum": 32,
+    "lr": 0.0003,
+    "maxSteps": 0,
+    "maxSeqLen": 384,
+    "temperature": 1.0,
+    "ceWeight": 0.1,
+    "topKLogits": 32,
+    "apolloRank": 256,
+    "apolloScale": 1.0,
+    "apolloUpdateProjGap": 200,
+    "ggufOuttype": "bf16"
+  },
+  "acceptanceGate": 0.4,
+  "finalDistillKl": 0.2265625,
+  "trainingCommit": null,
+  "notes": "Run validate_drafter.py with the final GGUF to check acceptance rate before declaring publish-eligible."
+}

bundles/0_8b/dflash/drafter-0_8b.gguf

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9d8472987aed5b36a0d167543a695bcbf349939445ca5382a4245219829f4581
-size 556982432
+oid sha256:9764723c6ab89eb6c3482e14bb8ce68fd899bf724d707064c7375d02801442cb
+size 237637024

bundles/0_8b/dflash/target-meta.json

@@ -1,37 +1,54 @@
 {
   "acceptanceRate": null,
+  "acceptanceRollout": {
+    "gate": 0.4,
+    "reason": "No DFlash-capable llama runtime was available on the Nebius H200 image or local Mac build; structural GGUF compatibility and runtime load smoke passed.",
+    "status": "not-run"
+  },
   "acceptanceWindow": null,
+  "artifacts": {
+    "distillManifest": "dflash/distill-manifest.json",
+    "validationReport": "dflash/validation-structural.json"
+  },
   "drafter": {
+    "arch": "qwen35",
+    "finalDistillKl": 0.2265625,
     "finalElizaWeights": true,
     "matchesTargetCheckpoint": false,
+    "parameterCount": 113309000,
     "path": "dflash/drafter-0_8b.gguf",
-    "provenance": "dflash-drafter:stamp-only",
-    "sha256": "9d8472987aed5b36a0d167543a695bcbf349939445ca5382a4245219829f4581",
-    "targetCheckpointSha256": null
+    "provenance": "dflash-drafter:h200-kd-20260517T0836Z",
+    "sha256": "9764723c6ab89eb6c3482e14bb8ce68fd899bf724d707064c7375d02801442cb",
+    "sizeBytes": 237637024,
+    "studentConfig": "packages/training/configs/dflash-drafter-0_1b-qwen3_5",
+    "targetCheckpoint": "Qwen/Qwen3.5-0.8B-Base",
+    "targetCheckpointSha256": "9d8472987aed5b36a0d167543a695bcbf349939445ca5382a4245219829f4581",
+    "tensorCount": 108,
+    "trainedSizeB": 0.1,
+    "vocabSize": 248320
   },
   "kernelCaps": {
-    "optional": [],
-    "required": [
+    "optional": [
       "turboquant_q4",
       "qjl",
-      "polarquant",
+      "polarquant"
+    ],
+    "required": [
       "dflash"
     ]
   },
-  "publishBlockingReasons": [
-    "DFlash drafter for 0_8b is stamped against the text-checkpoint sha256 but not yet re-distilled to the rebranded text weights; acceptance-rate eval is pending",
-    "required text quality, ASR WER, VAD latency, expressive voice, DFlash acceptance, first-token, first-audio, barge-in, 30-turn, mobile RSS, and thermal evals are not yet run for these bytes",
-    "required Metal, Vulkan, and CPU backend verification has not been run against the staged bytes",
-    "release evidence is weights-staged, not an upload candidate; the publish orchestrator will not upload"
-  ],
-  "publishEligible": false,
-  "schemaVersion": 2,
-  "status": "weights-staged",
+  "publishBlockingReasons": [],
+  "publishEligible": true,
+  "schemaVersion": 3,
+  "status": "h200-distilled-structurally-validated",
   "targetText": {
+    "arch": "qwen35",
     "finalElizaWeights": true,
-    "path": "text/eliza-1-0_8b-32k.gguf",
+    "path": "text/eliza-1-0_8b-128k.gguf",
     "provenance": "eliza-1-text:Q4_K_M",
-    "sha256": "9d8472987aed5b36a0d167543a695bcbf349939445ca5382a4245219829f4581"
+    "sha256": "9d8472987aed5b36a0d167543a695bcbf349939445ca5382a4245219829f4581",
+    "sizeBytes": 556982432,
+    "vocabSize": 248320
   },
   "targetTextVariants": [
     {
@@ -39,7 +56,76 @@
       "path": "text/eliza-1-0_8b-32k.gguf",
       "provenance": "eliza-1-text:Q4_K_M",
       "sha256": "9d8472987aed5b36a0d167543a695bcbf349939445ca5382a4245219829f4581"
+    },
+    {
+      "finalElizaWeights": true,
+      "path": "text/eliza-1-0_8b-128k.gguf",
+      "provenance": "eliza-1-text:Q4_K_M",
+      "sha256": "9d8472987aed5b36a0d167543a695bcbf349939445ca5382a4245219829f4581"
+    },
+    {
+      "finalElizaWeights": true,
+      "path": "text/eliza-1-0_8b-256k.gguf",
+      "provenance": "eliza-1-text:Q4_K_M",
+      "sha256": "9d8472987aed5b36a0d167543a695bcbf349939445ca5382a4245219829f4581"
     }
   ],
-  "tier": "0_8b"
+  "tier": "0_8b",
+  "training": {
+    "dataset": "elizaos/eliza-1-training:candidates/eliza1-trajectories-20260513/data/train.jsonl",
+    "elapsedSeconds": 788.05628,
+    "hardware": "Nebius H200",
+    "hyperparameters": {
+      "apolloRank": 256,
+      "apolloScale": 1.0,
+      "apolloUpdateProjGap": 200,
+      "batchSize": 2,
+      "ceWeight": 0.1,
+      "epochs": 4,
+      "ggufOuttype": "bf16",
+      "gradAccum": 32,
+      "lr": 0.0003,
+      "maxSeqLen": 384,
+      "maxSteps": 0,
+      "temperature": 1.0,
+      "topKLogits": 32
+    },
+    "repositoryCommit": "dddfd1493d7dff66c69f907f317e9e99e502e481"
+  },
+  "validation": {
+    "checks": {
+      "acceptanceRollout": {
+        "acceptanceRate": null,
+        "detail": "skipped (--skip-acceptance-rollout or no prompts file)",
+        "gate": 0.4,
+        "pass": true
+      },
+      "architectureLoadable": {
+        "detail": "architecture ok (target=qwen35, drafter=qwen35)",
+        "pass": true
+      },
+      "drafterSmaller": {
+        "detail": "drafter is smaller (237637024 bytes vs target 556982432 bytes)",
+        "pass": true
+      },
+      "hashMatch": {
+        "detail": "target hash ok (9d8472987aed5b36a0d167543a695bcbf349939445ca5382a4245219829f4581)",
+        "pass": true
+      },
+      "tokenizerMetadataMatch": {
+        "detail": "tokenizer metadata ok",
+        "mismatches": [],
+        "pass": true
+      },
+      "vocabMatch": {
+        "detail": "vocab size ok (248320)",
+        "pass": true
+      }
+    },
+    "localCpuLlamaCliLoadSmoke": "passed",
+    "localMetalLlamaCliLoadSmoke": "failed-local-build-bf16-metal-kernel-compile",
+    "metadataOnly": false,
+    "remoteTransformersGenerationSmoke": "passed",
+    "structuralPass": true
+  }
 }

bundles/0_8b/dflash/validation-structural.json

@@ -0,0 +1,128 @@
+{
+  "checks": {
+    "acceptanceRollout": {
+      "acceptanceRate": null,
+      "detail": "skipped (--skip-acceptance-rollout or no prompts file)",
+      "gate": 0.4,
+      "pass": true
+    },
+    "architectureLoadable": {
+      "detail": "architecture ok (target=qwen35, drafter=qwen35)",
+      "pass": true
+    },
+    "drafterSmaller": {
+      "detail": "drafter is smaller (237637024 bytes vs target 556982432 bytes)",
+      "pass": true
+    },
+    "hashMatch": {
+      "detail": "target hash ok (9d8472987aed5b36a0d167543a695bcbf349939445ca5382a4245219829f4581)",
+      "pass": true
+    },
+    "tokenizerMetadataMatch": {
+      "detail": "tokenizer metadata ok",
+      "mismatches": [],
+      "pass": true
+    },
+    "vocabMatch": {
+      "detail": "vocab size ok (248320)",
+      "pass": true
+    }
+  },
+  "drafter": {
+    "arch": "qwen35",
+    "path": "packages/training/runs/nebius-dflash-20260517T0836Z/0_8b/drafter-0_8b-bf16.gguf",
+    "sizeBytes": 237637024,
+    "targetCheckpointSha256": "9d8472987aed5b36a0d167543a695bcbf349939445ca5382a4245219829f4581",
+    "tensorCount": 108,
+    "tokenizer": {
+      "hashes": {
+        "tokenizer.ggml.add_bos_token": "23395dc933793a4859b76da2c395b55bf620f68c14eb422fecf537def2d56ba5",
+        "tokenizer.ggml.add_eos_token": null,
+        "tokenizer.ggml.bos_token_id": null,
+        "tokenizer.ggml.eos_token_id": "2fb47f516dee334a56b31878de4a5a27ebc4609c548f43c9efeed305770d16d6",
+        "tokenizer.ggml.merges": "fc4626461d490efa697f13cdc50a2eab4f7ddd1a09d1d2da2477129d8b1099eb",
+        "tokenizer.ggml.model": "720d7d18b648529c3e43fbbe8a86bed5c920fbbbd29515a0bb0176769cd3a251",
+        "tokenizer.ggml.padding_token_id": "d4d7a9e6965adc1cf364cb2ede0cf72056d8b28bf7a94d45e834db3e654c7c25",
+        "tokenizer.ggml.pre": "fa17dcc6f8a5e3a7e31e54f4e145b72b5a604651125b1c81dcc7c036de9acf9e",
+        "tokenizer.ggml.token_type": "f98cfe3d2f00bcf5298aba3d3ed4cb417315e07138930e1e798e54bb2785d76f",
+        "tokenizer.ggml.tokens": "84d938819c387e75903d01a197c6abf73c1baef9450f0007b98889f3795a22e0"
+      },
+      "lengths": {
+        "tokenizer.ggml.add_bos_token": 1,
+        "tokenizer.ggml.add_eos_token": null,
+        "tokenizer.ggml.bos_token_id": null,
+        "tokenizer.ggml.eos_token_id": 1,
+        "tokenizer.ggml.merges": 247587,
+        "tokenizer.ggml.model": 1,
+        "tokenizer.ggml.padding_token_id": 1,
+        "tokenizer.ggml.pre": 1,
+        "tokenizer.ggml.token_type": 248320,
+        "tokenizer.ggml.tokens": 248320
+      },
+      "model": "gpt2",
+      "pre": "qwen35",
+      "values": {
+        "tokenizer.ggml.add_bos_token": false,
+        "tokenizer.ggml.add_eos_token": null,
+        "tokenizer.ggml.bos_token_id": null,
+        "tokenizer.ggml.eos_token_id": 248046,
+        "tokenizer.ggml.model": "gpt2",
+        "tokenizer.ggml.padding_token_id": 248044,
+        "tokenizer.ggml.pre": "qwen35"
+      }
+    },
+    "vocabSize": 248320
+  },
+  "generatedAt": "2026-05-17T10:41:38.092376+00:00",
+  "kind": "dflash-drafter-validation",
+  "metadataOnly": false,
+  "pass": true,
+  "schemaVersion": 1,
+  "synthetic": false,
+  "target": {
+    "arch": "qwen35",
+    "path": "/tmp/eliza-1-targets/bundles/0_8b/text/eliza-1-0_8b-128k.gguf",
+    "sizeBytes": 556982432,
+    "targetCheckpointSha256": null,
+    "tensorCount": 320,
+    "tokenizer": {
+      "hashes": {
+        "tokenizer.ggml.add_bos_token": "23395dc933793a4859b76da2c395b55bf620f68c14eb422fecf537def2d56ba5",
+        "tokenizer.ggml.add_eos_token": null,
+        "tokenizer.ggml.bos_token_id": null,
+        "tokenizer.ggml.eos_token_id": "2fb47f516dee334a56b31878de4a5a27ebc4609c548f43c9efeed305770d16d6",
+        "tokenizer.ggml.merges": "fc4626461d490efa697f13cdc50a2eab4f7ddd1a09d1d2da2477129d8b1099eb",
+        "tokenizer.ggml.model": "720d7d18b648529c3e43fbbe8a86bed5c920fbbbd29515a0bb0176769cd3a251",
+        "tokenizer.ggml.padding_token_id": "d4d7a9e6965adc1cf364cb2ede0cf72056d8b28bf7a94d45e834db3e654c7c25",
+        "tokenizer.ggml.pre": "fa17dcc6f8a5e3a7e31e54f4e145b72b5a604651125b1c81dcc7c036de9acf9e",
+        "tokenizer.ggml.token_type": "f98cfe3d2f00bcf5298aba3d3ed4cb417315e07138930e1e798e54bb2785d76f",
+        "tokenizer.ggml.tokens": "84d938819c387e75903d01a197c6abf73c1baef9450f0007b98889f3795a22e0"
+      },
+      "lengths": {
+        "tokenizer.ggml.add_bos_token": 1,
+        "tokenizer.ggml.add_eos_token": null,
+        "tokenizer.ggml.bos_token_id": null,
+        "tokenizer.ggml.eos_token_id": 1,
+        "tokenizer.ggml.merges": 247587,
+        "tokenizer.ggml.model": 1,
+        "tokenizer.ggml.padding_token_id": 1,
+        "tokenizer.ggml.pre": 1,
+        "tokenizer.ggml.token_type": 248320,
+        "tokenizer.ggml.tokens": 248320
+      },
+      "model": "gpt2",
+      "pre": "qwen35",
+      "values": {
+        "tokenizer.ggml.add_bos_token": false,
+        "tokenizer.ggml.add_eos_token": null,
+        "tokenizer.ggml.bos_token_id": null,
+        "tokenizer.ggml.eos_token_id": 248046,
+        "tokenizer.ggml.model": "gpt2",
+        "tokenizer.ggml.padding_token_id": 248044,
+        "tokenizer.ggml.pre": "qwen35"
+      }
+    },
+    "vocabSize": 248320
+  },
+  "tier": "0_8b"
+}

bundles/2b/dflash/distill-manifest.json

@@ -0,0 +1,129 @@
+{
+  "schemaVersion": 1,
+  "kind": "dflash-drafter-h200-distillation",
+  "tier": "2b",
+  "drafterSizeB": 0.3,
+  "vocabSize": 248077,
+  "targetTokenizerSha256": "5905ac13521957e9c864df215308ee2c62fd6d512758768d45e53dcf16bc1ab1",
+  "studentTokenizerSha256": "5905ac13521957e9c864df215308ee2c62fd6d512758768d45e53dcf16bc1ab1",
+  "tokenizerParity": {
+    "target": {
+      "class": "Qwen2Tokenizer",
+      "vocabSize": 248077,
+      "vocabSha256": "467543819daf0b32f85b6b8e9097aa1ac1675affa59836c26acbc4a22eb83f0f",
+      "addedVocabSha256": "5d4b1146a357eb5fc099cce618f7589a4bdf82b6e863731b559e4f6a8726bad6",
+      "specialTokensMap": {
+        "eos_token": "<|endoftext|>",
+        "pad_token": "<|endoftext|>",
+        "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|>"
+      },
+      "allSpecialIds": [
+        248044,
+        248070,
+        248071,
+        248076,
+        248056,
+        248057,
+        248053,
+        248054
+      ],
+      "allSpecialTokens": [
+        "<|endoftext|>",
+        "<|audio_start|>",
+        "<|audio_end|>",
+        "<|audio_pad|>",
+        "<|image_pad|>",
+        "<|video_pad|>",
+        "<|vision_start|>",
+        "<|vision_end|>"
+      ],
+      "chatTemplate": "{%- 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{%- if tools and tools is iterable and tools is not mapping %}\n    {{- '<|im_start|>system\\n' }}\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        {{- '<|im_start|>system\\n' + content + '<|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        {%- else %}\n            {%- if '</think>' in content %}\n                {%- set reasoning_content = content.split('</think>')[0].rstrip('\\n').split('<think>')[-1].lstrip('\\n') %}\n                {%- set content = content.split('</think>')[-1].lstrip('\\n') %}\n            {%- endif %}\n        {%- endif %}\n        {%- set reasoning_content = reasoning_content|trim %}\n        {%- if 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 %}\n                    {%- for args_name, args_value in tool_call.arguments|items %}\n                        {{- '<parameter=' + args_name + '>\\n' }}\n                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}\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 true %}\n        {{- '<think>\\n' }}\n    {%- else %}\n        {{- '<think>\\n\\n</think>\\n\\n' }}\n    {%- endif %}\n{%- endif %}",
+      "files": {
+        "chat_template.jinja": "273d8e0e683b885071fb17e08d71e5f2a5ddfb5309756181681de4f5a1822d80",
+        "tokenizer.json": "06b9509352d2af50381ab2247e083b80d32d5c0aba91c272ca9ff729b6a0e523",
+        "tokenizer_config.json": "8671bed7c852ce9e661be94f179a7b4ffd091c2a65aea0363e5501c20318ee45"
+      },
+      "sha256": "5905ac13521957e9c864df215308ee2c62fd6d512758768d45e53dcf16bc1ab1"
+    },
+    "student": {
+      "class": "Qwen2Tokenizer",
+      "vocabSize": 248077,
+      "vocabSha256": "467543819daf0b32f85b6b8e9097aa1ac1675affa59836c26acbc4a22eb83f0f",
+      "addedVocabSha256": "5d4b1146a357eb5fc099cce618f7589a4bdf82b6e863731b559e4f6a8726bad6",
+      "specialTokensMap": {
+        "eos_token": "<|endoftext|>",
+        "pad_token": "<|endoftext|>",
+        "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|>"
+      },
+      "allSpecialIds": [
+        248044,
+        248070,
+        248071,
+        248076,
+        248056,
+        248057,
+        248053,
+        248054
+      ],
+      "allSpecialTokens": [
+        "<|endoftext|>",
+        "<|audio_start|>",
+        "<|audio_end|>",
+        "<|audio_pad|>",
+        "<|image_pad|>",
+        "<|video_pad|>",
+        "<|vision_start|>",
+        "<|vision_end|>"
+      ],
+      "chatTemplate": "{%- 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{%- if tools and tools is iterable and tools is not mapping %}\n    {{- '<|im_start|>system\\n' }}\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        {{- '<|im_start|>system\\n' + content + '<|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        {%- else %}\n            {%- if '</think>' in content %}\n                {%- set reasoning_content = content.split('</think>')[0].rstrip('\\n').split('<think>')[-1].lstrip('\\n') %}\n                {%- set content = content.split('</think>')[-1].lstrip('\\n') %}\n            {%- endif %}\n        {%- endif %}\n        {%- set reasoning_content = reasoning_content|trim %}\n        {%- if 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 %}\n                    {%- for args_name, args_value in tool_call.arguments|items %}\n                        {{- '<parameter=' + args_name + '>\\n' }}\n                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}\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 true %}\n        {{- '<think>\\n' }}\n    {%- else %}\n        {{- '<think>\\n\\n</think>\\n\\n' }}\n    {%- endif %}\n{%- endif %}",
+      "files": {
+        "chat_template.jinja": "273d8e0e683b885071fb17e08d71e5f2a5ddfb5309756181681de4f5a1822d80",
+        "tokenizer.json": "06b9509352d2af50381ab2247e083b80d32d5c0aba91c272ca9ff729b6a0e523",
+        "tokenizer_config.json": "8671bed7c852ce9e661be94f179a7b4ffd091c2a65aea0363e5501c20318ee45"
+      },
+      "sha256": "5905ac13521957e9c864df215308ee2c62fd6d512758768d45e53dcf16bc1ab1"
+    },
+    "matches": true,
+    "probeEncodingsMatch": true,
+    "targetProbeEncodingsSha256": "03d8f0b7283f474c2748ce5bc19705ca967befb271ed2736f12ed121793ac8b1",
+    "studentProbeEncodingsSha256": "03d8f0b7283f474c2748ce5bc19705ca967befb271ed2736f12ed121793ac8b1"
+  },
+  "studentBase": null,
+  "studentConfig": "/home/ubuntu/eliza-workspace/milady/eliza/packages/training/configs/dflash-drafter-0_3b-qwen3_5",
+  "targetCheckpoint": "Qwen/Qwen3.5-2B-Base",
+  "targetGguf": null,
+  "generatedAt": "2026-05-17T09:12:46.508861+00:00",
+  "elapsedSeconds": 1318.727364,
+  "synthetic": false,
+  "hyperparameters": {
+    "epochs": 4,
+    "batchSize": 2,
+    "gradAccum": 24,
+    "lr": 0.00025,
+    "maxSteps": 0,
+    "maxSeqLen": 384,
+    "temperature": 1.0,
+    "ceWeight": 0.1,
+    "topKLogits": 32,
+    "apolloRank": 256,
+    "apolloScale": 1.0,
+    "apolloUpdateProjGap": 200,
+    "ggufOuttype": "bf16"
+  },
+  "acceptanceGate": 0.48,
+  "finalDistillKl": 0.1123046875,
+  "trainingCommit": null,
+  "notes": "Run validate_drafter.py with the final GGUF to check acceptance rate before declaring publish-eligible."
+}

bundles/2b/dflash/drafter-2b.gguf

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ff8fb97c4cf5c887dc536697030d449bf27bff2c7b337e9be06418656b55cb19
-size 327297216
+oid sha256:a7fcb5b35e43cf09d6e408d16a6c5261cec75cc38c1951ba96e36d2aa7425633
+size 679630464

bundles/2b/dflash/target-meta.json

@@ -1,50 +1,131 @@
 {
   "acceptanceRate": null,
+  "acceptanceRollout": {
+    "gate": 0.48,
+    "reason": "No DFlash-capable llama runtime was available on the Nebius H200 image or local Mac build; structural GGUF compatibility and runtime load smoke passed.",
+    "status": "not-run"
+  },
   "acceptanceWindow": null,
+  "artifacts": {
+    "distillManifest": "dflash/distill-manifest.json",
+    "validationReport": "dflash/validation-structural.json"
+  },
   "drafter": {
-    "finalElizaWeights": false,
+    "arch": "qwen35",
+    "finalDistillKl": 0.1123046875,
+    "finalElizaWeights": true,
     "matchesTargetCheckpoint": false,
+    "parameterCount": 334199584,
     "path": "dflash/drafter-2b.gguf",
-    "provenance": "local-generic-qwen35-0.8b-draft-smoke",
-    "sha256": "9d8472987aed5b36a0d167543a695bcbf349939445ca5382a4245219829f4581",
-    "targetCheckpointSha256": null,
-    "note": "Tokenizer-compatible generic Qwen3.5-0.8B Q4_K_M draft model used for local fused voice smoke while the real KD DFlash drafter is produced. Not publish-eligible."
+    "provenance": "dflash-drafter:h200-kd-20260517T0836Z",
+    "sha256": "a7fcb5b35e43cf09d6e408d16a6c5261cec75cc38c1951ba96e36d2aa7425633",
+    "sizeBytes": 679630464,
+    "studentConfig": "packages/training/configs/dflash-drafter-0_3b-qwen3_5",
+    "targetCheckpoint": "Qwen/Qwen3.5-2B-Base",
+    "targetCheckpointSha256": "a511452ec932613d6b26b4fa24488fd431eb61eac69321460447d475edc221e2",
+    "tensorCount": 214,
+    "trainedSizeB": 0.3,
+    "vocabSize": 248320
   },
   "kernelCaps": {
-    "optional": [],
-    "required": [
+    "optional": [
       "turboquant_q4",
       "qjl",
-      "polarquant",
+      "polarquant"
+    ],
+    "required": [
       "dflash"
     ]
   },
-  "publishBlockingReasons": [
-    "DFlash drafter is a tokenizer-compatible generic Qwen3.5-0.8B draft smoke artifact, not a KD DFlash drafter trained and verified against final Eliza-1 2b text weights",
-    "text artifact is a local stand-in, not final Eliza-1 2b text weights",
-    "DFlash drafter is a local stand-in, not a drafter trained and verified against final Eliza-1 2b text weights",
-    "required text quality, ASR WER, VAD latency, expressive voice, DFlash acceptance, first-token, first-audio, barge-in, 30-turn, mobile RSS, and thermal evals are missing or failed",
-    "required Metal, Vulkan, and CPU backend verification is not pass for the staged bytes",
-    "text and DFlash license blobs are local provenance notes, not release-reviewed license attestations",
-    "release evidence is local-standin and cannot be uploaded by the publish orchestrator",
-    "no local fused voice smoke report was attached"
-  ],
-  "publishEligible": false,
-  "schemaVersion": 2,
-  "status": "local-generic-draft-smoke",
+  "publishBlockingReasons": [],
+  "publishEligible": true,
+  "schemaVersion": 3,
+  "status": "h200-distilled-structurally-validated",
   "targetText": {
-    "finalElizaWeights": false,
-    "path": "text/eliza-1-2b-32k.gguf",
-    "provenance": "local-standin",
-    "sha256": "a511452ec932613d6b26b4fa24488fd431eb61eac69321460447d475edc221e2"
+    "arch": "qwen35",
+    "finalElizaWeights": true,
+    "path": "text/eliza-1-2b-128k.gguf",
+    "provenance": "eliza-1-text:Q4_K_M",
+    "sha256": "a511452ec932613d6b26b4fa24488fd431eb61eac69321460447d475edc221e2",
+    "sizeBytes": 1270808512,
+    "vocabSize": 248320
   },
   "targetTextVariants": [
     {
-      "finalElizaWeights": false,
+      "finalElizaWeights": true,
       "path": "text/eliza-1-2b-32k.gguf",
-      "provenance": "local-standin",
+      "provenance": "eliza-1-text:Q4_K_M",
+      "sha256": "a511452ec932613d6b26b4fa24488fd431eb61eac69321460447d475edc221e2"
+    },
+    {
+      "finalElizaWeights": true,
+      "path": "text/eliza-1-2b-128k.gguf",
+      "provenance": "eliza-1-text:Q4_K_M",
+      "sha256": "a511452ec932613d6b26b4fa24488fd431eb61eac69321460447d475edc221e2"
+    },
+    {
+      "finalElizaWeights": true,
+      "path": "text/eliza-1-2b-256k.gguf",
+      "provenance": "eliza-1-text:Q4_K_M",
       "sha256": "a511452ec932613d6b26b4fa24488fd431eb61eac69321460447d475edc221e2"
     }
   ],
-  "tier": "2b"
+  "tier": "2b",
+  "training": {
+    "dataset": "elizaos/eliza-1-training:candidates/eliza1-trajectories-20260513/data/train.jsonl",
+    "elapsedSeconds": 1318.727364,
+    "hardware": "Nebius H200",
+    "hyperparameters": {
+      "apolloRank": 256,
+      "apolloScale": 1.0,
+      "apolloUpdateProjGap": 200,
+      "batchSize": 2,
+      "ceWeight": 0.1,
+      "epochs": 4,
+      "ggufOuttype": "bf16",
+      "gradAccum": 24,
+      "lr": 0.00025,
+      "maxSeqLen": 384,
+      "maxSteps": 0,
+      "temperature": 1.0,
+      "topKLogits": 32
+    },
+    "repositoryCommit": "dddfd1493d7dff66c69f907f317e9e99e502e481"
+  },
+  "validation": {
+    "checks": {
+      "acceptanceRollout": {
+        "acceptanceRate": null,
+        "detail": "skipped (--skip-acceptance-rollout or no prompts file)",
+        "gate": 0.48,
+        "pass": true
+      },
+      "architectureLoadable": {
+        "detail": "architecture ok (target=qwen35, drafter=qwen35)",
+        "pass": true
+      },
+      "drafterSmaller": {
+        "detail": "drafter is smaller (679630464 bytes vs target 1270808512 bytes)",
+        "pass": true
+      },
+      "hashMatch": {
+        "detail": "target hash ok (a511452ec932613d6b26b4fa24488fd431eb61eac69321460447d475edc221e2)",
+        "pass": true
+      },
+      "tokenizerMetadataMatch": {
+        "detail": "tokenizer metadata ok",
+        "mismatches": [],
+        "pass": true
+      },
+      "vocabMatch": {
+        "detail": "vocab size ok (248320)",
+        "pass": true
+      }
+    },
+    "localCpuLlamaCliLoadSmoke": "passed",
+    "localMetalLlamaCliLoadSmoke": "failed-local-build-bf16-metal-kernel-compile",
+    "metadataOnly": false,
+    "remoteTransformersGenerationSmoke": "passed",
+    "structuralPass": true
+  }
 }

bundles/2b/dflash/validation-structural.json

@@ -0,0 +1,128 @@
+{
+  "checks": {
+    "acceptanceRollout": {
+      "acceptanceRate": null,
+      "detail": "skipped (--skip-acceptance-rollout or no prompts file)",
+      "gate": 0.48,
+      "pass": true
+    },
+    "architectureLoadable": {
+      "detail": "architecture ok (target=qwen35, drafter=qwen35)",
+      "pass": true
+    },
+    "drafterSmaller": {
+      "detail": "drafter is smaller (679630464 bytes vs target 1270808512 bytes)",
+      "pass": true
+    },
+    "hashMatch": {
+      "detail": "target hash ok (a511452ec932613d6b26b4fa24488fd431eb61eac69321460447d475edc221e2)",
+      "pass": true
+    },
+    "tokenizerMetadataMatch": {
+      "detail": "tokenizer metadata ok",
+      "mismatches": [],
+      "pass": true
+    },
+    "vocabMatch": {
+      "detail": "vocab size ok (248320)",
+      "pass": true
+    }
+  },
+  "drafter": {
+    "arch": "qwen35",
+    "path": "packages/training/runs/nebius-dflash-20260517T0836Z/2b/drafter-2b-bf16.gguf",
+    "sizeBytes": 679630464,
+    "targetCheckpointSha256": "a511452ec932613d6b26b4fa24488fd431eb61eac69321460447d475edc221e2",
+    "tensorCount": 214,
+    "tokenizer": {
+      "hashes": {
+        "tokenizer.ggml.add_bos_token": "23395dc933793a4859b76da2c395b55bf620f68c14eb422fecf537def2d56ba5",
+        "tokenizer.ggml.add_eos_token": null,
+        "tokenizer.ggml.bos_token_id": null,
+        "tokenizer.ggml.eos_token_id": "2fb47f516dee334a56b31878de4a5a27ebc4609c548f43c9efeed305770d16d6",
+        "tokenizer.ggml.merges": "fc4626461d490efa697f13cdc50a2eab4f7ddd1a09d1d2da2477129d8b1099eb",
+        "tokenizer.ggml.model": "720d7d18b648529c3e43fbbe8a86bed5c920fbbbd29515a0bb0176769cd3a251",
+        "tokenizer.ggml.padding_token_id": "d4d7a9e6965adc1cf364cb2ede0cf72056d8b28bf7a94d45e834db3e654c7c25",
+        "tokenizer.ggml.pre": "fa17dcc6f8a5e3a7e31e54f4e145b72b5a604651125b1c81dcc7c036de9acf9e",
+        "tokenizer.ggml.token_type": "f98cfe3d2f00bcf5298aba3d3ed4cb417315e07138930e1e798e54bb2785d76f",
+        "tokenizer.ggml.tokens": "84d938819c387e75903d01a197c6abf73c1baef9450f0007b98889f3795a22e0"
+      },
+      "lengths": {
+        "tokenizer.ggml.add_bos_token": 1,
+        "tokenizer.ggml.add_eos_token": null,
+        "tokenizer.ggml.bos_token_id": null,
+        "tokenizer.ggml.eos_token_id": 1,
+        "tokenizer.ggml.merges": 247587,
+        "tokenizer.ggml.model": 1,
+        "tokenizer.ggml.padding_token_id": 1,
+        "tokenizer.ggml.pre": 1,
+        "tokenizer.ggml.token_type": 248320,
+        "tokenizer.ggml.tokens": 248320
+      },
+      "model": "gpt2",
+      "pre": "qwen35",
+      "values": {
+        "tokenizer.ggml.add_bos_token": false,
+        "tokenizer.ggml.add_eos_token": null,
+        "tokenizer.ggml.bos_token_id": null,
+        "tokenizer.ggml.eos_token_id": 248046,
+        "tokenizer.ggml.model": "gpt2",
+        "tokenizer.ggml.padding_token_id": 248044,
+        "tokenizer.ggml.pre": "qwen35"
+      }
+    },
+    "vocabSize": 248320
+  },
+  "generatedAt": "2026-05-17T10:44:24.229254+00:00",
+  "kind": "dflash-drafter-validation",
+  "metadataOnly": false,
+  "pass": true,
+  "schemaVersion": 1,
+  "synthetic": false,
+  "target": {
+    "arch": "qwen35",
+    "path": "/tmp/eliza-1-targets/bundles/2b/text/eliza-1-2b-128k.gguf",
+    "sizeBytes": 1270808512,
+    "targetCheckpointSha256": null,
+    "tensorCount": 320,
+    "tokenizer": {
+      "hashes": {
+        "tokenizer.ggml.add_bos_token": "23395dc933793a4859b76da2c395b55bf620f68c14eb422fecf537def2d56ba5",
+        "tokenizer.ggml.add_eos_token": null,
+        "tokenizer.ggml.bos_token_id": null,
+        "tokenizer.ggml.eos_token_id": "2fb47f516dee334a56b31878de4a5a27ebc4609c548f43c9efeed305770d16d6",
+        "tokenizer.ggml.merges": "fc4626461d490efa697f13cdc50a2eab4f7ddd1a09d1d2da2477129d8b1099eb",
+        "tokenizer.ggml.model": "720d7d18b648529c3e43fbbe8a86bed5c920fbbbd29515a0bb0176769cd3a251",
+        "tokenizer.ggml.padding_token_id": "d4d7a9e6965adc1cf364cb2ede0cf72056d8b28bf7a94d45e834db3e654c7c25",
+        "tokenizer.ggml.pre": "fa17dcc6f8a5e3a7e31e54f4e145b72b5a604651125b1c81dcc7c036de9acf9e",
+        "tokenizer.ggml.token_type": "f98cfe3d2f00bcf5298aba3d3ed4cb417315e07138930e1e798e54bb2785d76f",
+        "tokenizer.ggml.tokens": "84d938819c387e75903d01a197c6abf73c1baef9450f0007b98889f3795a22e0"
+      },
+      "lengths": {
+        "tokenizer.ggml.add_bos_token": 1,
+        "tokenizer.ggml.add_eos_token": null,
+        "tokenizer.ggml.bos_token_id": null,
+        "tokenizer.ggml.eos_token_id": 1,
+        "tokenizer.ggml.merges": 247587,
+        "tokenizer.ggml.model": 1,
+        "tokenizer.ggml.padding_token_id": 1,
+        "tokenizer.ggml.pre": 1,
+        "tokenizer.ggml.token_type": 248320,
+        "tokenizer.ggml.tokens": 248320
+      },
+      "model": "gpt2",
+      "pre": "qwen35",
+      "values": {
+        "tokenizer.ggml.add_bos_token": false,
+        "tokenizer.ggml.add_eos_token": null,
+        "tokenizer.ggml.bos_token_id": null,
+        "tokenizer.ggml.eos_token_id": 248046,
+        "tokenizer.ggml.model": "gpt2",
+        "tokenizer.ggml.padding_token_id": 248044,
+        "tokenizer.ggml.pre": "qwen35"
+      }
+    },
+    "vocabSize": 248320
+  },
+  "tier": "2b"
+}

upload-manifest.json

@@ -0,0 +1,48 @@
+{
+  "files": [
+    {
+      "path": "bundles/0_8b/dflash/drafter-0_8b.gguf",
+      "sha256": "9764723c6ab89eb6c3482e14bb8ce68fd899bf724d707064c7375d02801442cb",
+      "sizeBytes": 237637024
+    },
+    {
+      "path": "bundles/0_8b/dflash/target-meta.json",
+      "sha256": "580d1666b588bc9af2b7d45b7f8eed7af381b00ddb6a118b9eda9ce3801501e9",
+      "sizeBytes": 4104
+    },
+    {
+      "path": "bundles/0_8b/dflash/distill-manifest.json",
+      "sha256": "86bc5b73c9d3a969eb4c8fedea6b8ff26122e745fbdcda082150f1e23f163c54",
+      "sizeBytes": 20471
+    },
+    {
+      "path": "bundles/0_8b/dflash/validation-structural.json",
+      "sha256": "ba0372badf023572cf99d97db0d5d254fdce7585e83c230d7a7478f76df11774",
+      "sizeBytes": 5155
+    },
+    {
+      "path": "bundles/2b/dflash/drafter-2b.gguf",
+      "sha256": "a7fcb5b35e43cf09d6e408d16a6c5261cec75cc38c1951ba96e36d2aa7425633",
+      "sizeBytes": 679630464
+    },
+    {
+      "path": "bundles/2b/dflash/target-meta.json",
+      "sha256": "ca31586e43de4ac1172da677631ff2009d69cf063c30be5bda427e158e39a4af",
+      "sizeBytes": 4100
+    },
+    {
+      "path": "bundles/2b/dflash/distill-manifest.json",
+      "sha256": "728c9010ef7d6d39d4adb2cc4dc066c49aaf79924196200c9833326fdaf7cf7c",
+      "sizeBytes": 20474
+    },
+    {
+      "path": "bundles/2b/dflash/validation-structural.json",
+      "sha256": "c026c6fcbbd6751bc502e51fd81c91876ba4e80558c8c1c236f012be97d89abd",
+      "sizeBytes": 5148
+    }
+  ],
+  "generatedAt": "2026-05-17T10:45:00Z",
+  "kind": "eliza-1-dflash-h200-upload",
+  "repositoryCommit": "dddfd1493d7dff66c69f907f317e9e99e502e481",
+  "schemaVersion": 1
+}