初始化项目，由ModelHub XC社区提供模型

Model: HarleyCooper/Qwen3-0.6B-Dakota-Grammar-RL-400
Source: Original Platform

.gitattributes

@@ -0,0 +1,36 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text
+tokenizer.json filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,241 @@
+---
+license: apache-2.0
+base_model: Qwen/Qwen3-0.6B
+tags:
+  - reinforcement-learning
+  - rl
+  - dakota-language
+  - grammar
+  - composition-rewards
+  - non-coding
+  - qualitative-tasks
+  - grpo
+  - prime-intellect
+  - verifiers
+language:
+  - en
+  - dak
+pipeline_tag: text-generation
+---
+
+# Qwen3-0.6B-Dakota-Grammar-RL-400
+
+<div align="center">
+
+<img src="https://raw.githubusercontent.com/HarleyCoops/Dakota1890/main/Public/Prepositions.jpg" alt="Dakota Prepositions - High Detail Scan" style="width: 100%; max-width: 1200px; height: auto; border-radius: 8px;">
+
+*Exceptional level of detail preserved from the 1890 source material — every character, accent, and linguistic nuance captured with precision*
+
+</div>
+
+## Model Description
+
+This model is a reinforcement learning (RL) fine-tuned version of `Qwen/Qwen3-0.6B`, trained specifically for Dakota language grammar and translation tasks using **GRPO (Group Relative Policy Optimization) with compositional reward functions on qualitative linguistic tasks**. 
+
+**GRPO is effective for linguistic-structure learning when qualitative goals are expressed as verifiable, compositional rewards.**
+
+### Key Features
+
+- **GRPO for Linguistic Structure**: GRPO is effective for linguistic-structure learning when qualitative goals are expressed as verifiable, compositional rewards
+- **Compositional Rewards**: Multi-component reward function combining character preservation (40%), morphological accuracy (40%), and semantic correctness (20%)
+- **Rapid Learning**: 150.3% improvement in 400 steps, with 90% of improvement achieved in just 21% of training
+- **Dakota Language Focus**: Trained on 5,657 grammar tasks extracted from the 1890 Dakota-English Dictionary
+- **Special Character Preservation**: Maintains Dakota orthography (ć, š, ŋ, ḣ, ṡ, á, é, í, ó, ú, etc.)
+- **Stable Training**: Low unmasked KL (0.092) demonstrates no catastrophic forgetting
+
+<div align="center">
+
+**Complete project repository with all code, data, and training traces:**
+
+[https://github.com/HarleyCoops/Dakota1890](https://github.com/HarleyCoops/Dakota1890)
+
+</div>
+
+## Training Details
+
+### Training Data
+
+- **Source**: 1890 Dakota-English Dictionary grammar section (pages 31-92)
+- **Tasks**: 5,657 training tasks covering:
+  - Morphology (affix application, word formation)
+  - Translation (Dakota ↔ English)
+  - Reverse translation
+  - Syntax (sentence structure)
+  - Pattern identification
+- **Difficulty Levels**: Easy (1,998), Medium (2,155), Hard (398), Advanced (1,106)
+
+<div align="center" style="margin: 2rem 0;">
+
+<img src="https://raw.githubusercontent.com/HarleyCoops/Dakota1890/main/Public/grammar.jpg" alt="Dakota Grammar - Historical Text Detail" style="width: 100%; max-width: 1200px; height: auto; border-radius: 8px;">
+
+*Grammar section from the 1890 Dakota-English Dictionary showing detailed linguistic rules and interlinear text*
+
+</div>
+
+<div align="center" style="margin: 2rem 0;">
+
+<img src="https://raw.githubusercontent.com/HarleyCoops/Dakota1890/main/Public/dictionary2.jpg" alt="Dakota Dictionary - Historical Text Detail" style="width: 100%; max-width: 1200px; height: auto; border-radius: 8px;">
+
+*Dictionary entries from the 1890 source material, preserving Dakota orthography and special characters*
+
+</div>
+
+### Training Procedure
+
+- **Framework**: Prime Intellect RL (prime-rl)
+- **Algorithm**: GRPO (Group Relative Policy Optimization)
+- **Base Model**: Qwen/Qwen3-0.6B (small instruct model optimized for RL)
+- **Training Steps**: 400 steps (all completed)
+- **Total Samples**: 102,400 samples processed
+- **Batch Size**: 256
+- **Sequence Length**: 1,536 tokens
+- **Rollouts per Example**: 8
+- **Learning Rate**: 1e-6
+- **Checkpoint Interval**: Every 100 steps (kept 3 most recent)
+- **GPUs**: 
+  - Trainer: GPU 0
+  - Inference: GPU 0
+
+### Reward Function Composition
+
+The model was trained using a **compositional reward function** that decomposes qualitative linguistic tasks into verifiable quantitative components:
+
+1. **Character Preservation (40% weight)**: Verifiable Unicode-level correctness for Dakota special characters (ć, š, ŋ, ḣ, ṡ, á, é, í, ó, ú)
+2. **Morphological Accuracy (40% weight)**: Pattern-matching against grammar rules for affix application and word formation
+3. **Semantic Correctness (20% weight)**: Meaning preservation metrics for translation quality
+
+**Why This Matters**: By decomposing rewards into independently verifiable components, we transform qualitative tasks (traditionally considered unsuitable for RL) into quantitatively optimizable objectives. This enables GRPO to work effectively because:
+- Each component is independently verifiable (no human judgment needed)
+- Gradients flow through each component (model learns what to prioritize)
+- Multi-dimensional feedback (model knows exactly what it got wrong)
+
+### Environment
+
+- **Environment**: `dakota_grammar_translation` (local installation)
+- **Framework**: Verifiers-compatible RL environment
+- **Parser**: DakotaTranslationParser (preserves Dakota orthography)
+
+## Training Results
+
+### Key Achievements
+
+- **150.3% improvement** in overall reward (0.128 → 0.321, peak: 0.345)
+- **Rapid learning**: 90% of improvement achieved in first 85 steps (21.25% of training)
+- **Sample efficiency**: 0.000483 improvement per step - demonstrating dense learning signals
+- **Stable training**: Controlled KL divergence with unmasked KL remaining low (mean: 0.094, final: 0.092)
+- **Policy confidence**: Entropy decreased from 0.93 to 0.28, showing increased model certainty
+
+### Training Metrics
+
+- **Final Entropy**: 0.28 (mean), 0.024 (median)
+- **Inference Probabilities**: Increased throughout training
+- **Peak Memory**: 13.9 GiB
+- **KL Divergence**: 
+  - Masked KL: 11.96 (final) - substantial policy adaptation for Dakota-specific tokens
+  - Unmasked KL: 0.092 (final) - preserved general language capabilities
+  - Overall KL: 5.03 (final) - controlled policy adaptation
+
+### W&B Runs
+
+- **Project**: dakota-rl-grammar
+- **Trainer Run**: [`yut26kcm`](https://wandb.ai/christian-cooper-us/dakota-rl-grammar/runs/yut26kcm) - `dakota-0.6b-ledger-test-400-trainer`
+- **Orchestrator Run**: [`1y33h9zr`](https://wandb.ai/christian-cooper-us/dakota-rl-grammar/runs/1y33h9zr) - `dakota-0.6b-ledger-test-400-orchestrator`
+
+### Training Visualizations
+
+![Comprehensive Dashboard](https://raw.githubusercontent.com/HarleyCoops/Dakota1890/main/wandb_visualizations/comprehensive_dashboard.png)
+
+*Comprehensive dashboard showing reward progression, component performance, loss dynamics, entropy, and KL divergence*
+
+![Reward Progression](https://raw.githubusercontent.com/HarleyCoops/Dakota1890/main/wandb_visualizations/reward_progression.png)
+
+*Reward progression demonstrating 150.3% improvement with 90% achieved in just 21% of training*
+
+![Training Metrics](https://raw.githubusercontent.com/HarleyCoops/Dakota1890/main/wandb_visualizations/training_metrics.png)
+
+*Training metrics showing stable optimization, decreasing entropy, and controlled KL divergence*
+
+![Performance Metrics](https://raw.githubusercontent.com/HarleyCoops/Dakota1890/main/wandb_visualizations/performance_metrics.png)
+
+*Performance metrics showing consistent throughput and GPU utilization*
+
+## GRPO for Qualitative Tasks: Significance
+
+**GRPO is effective for linguistic-structure learning when qualitative goals are expressed as verifiable, compositional rewards.** This is significant because:
+
+### Why This Matters
+
+GRPO has been successfully applied to **quantitative domains** (code generation, mathematical reasoning) where correctness is verifiable and rewards are clear. However, **qualitative tasks** like language learning, translation, and grammar have traditionally been considered unsuitable for RL because:
+
+1. **Subjective evaluation**: "Is this translation good?" lacks clear criteria
+2. **Multi-dimensional quality**: A translation can be semantically correct but orthographically wrong
+3. **Nuanced feedback**: Binary correct/incorrect fails to capture partial correctness
+
+### Our Solution: Compositional Rewards
+
+By decomposing rewards into **linguistic primitives** (character preservation, morphological accuracy, semantic correctness), we transform qualitative tasks into **quantitatively optimizable objectives**. This decomposition enables GRPO to work effectively because each component is independently verifiable, gradients flow through each component, and the model receives multi-dimensional feedback.
+
+### Key Results Demonstrating Significance
+
+1. **150.3% improvement in 400 steps** - Comparable to GRPO performance on coding tasks
+2. **90% improvement in 21% of training** - Demonstrates dense learning signals from compositional rewards
+3. **Low unmasked KL (0.092)** - Model specializes without catastrophic forgetting
+4. **Stable training dynamics** - No reward hacking or instability issues
+
+### Implications
+
+**GRPO is effective for linguistic-structure learning when qualitative goals are expressed as verifiable, compositional rewards.** When qualitative tasks are decomposed into verifiable components, they become as learnable as coding or math. This opens new possibilities for:
+
+- **Low-resource language learning** (this work)
+- **Style transfer** (decompose into syntax, semantics, register)
+- **Dialogue systems** (decompose into coherence, relevance, appropriateness)
+- **Creative tasks** (decompose into structure, originality, coherence)
+
+## Intended Use
+
+This model is intended for:
+- Research on GRPO for qualitative linguistic tasks
+- Demonstrating compositional reward functions in RL pipelines
+- Dakota language grammar and translation tasks
+- Testing RL effectiveness on linguistic-structure learning with compositional rewards
+- Low-resource language learning applications
+
+## Limitations
+
+- Small model size (0.6B parameters) limits capacity for complex grammar rules
+- Trained on historical dictionary data (1890) which may not reflect modern Dakota usage
+- Limited to single-turn and multi-turn chat formats
+- Requires Dakota language knowledge for proper evaluation
+- 400-step training run (test run) - longer training may yield further improvements
+
+## Ethical Considerations
+
+- Trained on historical linguistic data from indigenous language documentation
+- Should be used respectfully and in consultation with Dakota language communities
+- Not intended to replace human language experts or native speakers
+- Part of language preservation and revitalization efforts
+
+## Citation
+
+```bibtex
+@misc{dakota1890-rl-400-2024,
+  title={Qwen3-0.6B-Dakota-Grammar-RL-400: GRPO for Qualitative Linguistic Tasks with Compositional Rewards},
+  author={Christian H. Cooper},
+  year={2024},
+  url={https://huggingface.co/HarleyCooper/Qwen3-0.6B-Dakota-Grammar-RL-400},
+  note={Demonstrates GRPO effectiveness on qualitative tasks through compositional reward decomposition}
+}
+```
+
+## Acknowledgments
+
+- Base model: Qwen/Qwen3-0.6B by Alibaba Cloud
+- Training framework: Prime Intellect RL (prime-rl)
+- Source material: 1890 Dakota-English Dictionary by Stephen Return Riggs
+- Environment: Dakota1890 RL environment (dakota_grammar_translation)
+- Weights & Biases: Training monitoring and visualization
+
+## Model Card Contact
+
+For questions or issues, please contact: Raise an Issue in the [Repository](https://github.com/HarleyCoops/Dakota1890)
+

added_tokens.json

@@ -0,0 +1,28 @@
+{
+  "</think>": 151668,
+  "</tool_call>": 151658,
+  "</tool_response>": 151666,
+  "<think>": 151667,
+  "<tool_call>": 151657,
+  "<tool_response>": 151665,
+  "<|box_end|>": 151649,
+  "<|box_start|>": 151648,
+  "<|endoftext|>": 151643,
+  "<|file_sep|>": 151664,
+  "<|fim_middle|>": 151660,
+  "<|fim_pad|>": 151662,
+  "<|fim_prefix|>": 151659,
+  "<|fim_suffix|>": 151661,
+  "<|im_end|>": 151645,
+  "<|im_start|>": 151644,
+  "<|image_pad|>": 151655,
+  "<|object_ref_end|>": 151647,
+  "<|object_ref_start|>": 151646,
+  "<|quad_end|>": 151651,
+  "<|quad_start|>": 151650,
+  "<|repo_name|>": 151663,
+  "<|video_pad|>": 151656,
+  "<|vision_end|>": 151653,
+  "<|vision_pad|>": 151654,
+  "<|vision_start|>": 151652
+}

chat_template.jinja

@@ -0,0 +1,89 @@
+{%- if tools %}
+    {{- '<|im_start|>system\n' }}
+    {%- if messages[0].role == 'system' %}
+        {{- messages[0].content + '\n\n' }}
+    {%- endif %}
+    {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" and message.content is string and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}
+        {%- set ns.multi_step_tool = false %}
+        {%- set ns.last_query_index = index %}
+    {%- endif %}
+{%- endfor %}
+{%- for message in messages %}
+    {%- if message.content is string %}
+        {%- set content = message.content %}
+    {%- else %}
+        {%- set content = '' %}
+    {%- endif %}
+    {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {%- if loop.last or (not loop.last and reasoning_content) %}
+                {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
+            {%- else %}
+                {{- '<|im_start|>' + message.role + '\n' + content }}
+            {%- endif %}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if (loop.first and content) or (not loop.first) %}
+                    {{- '\n' }}
+                {%- endif %}
+                {%- if tool_call.function %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {{- '<tool_call>\n{"name": "' }}
+                {{- tool_call.name }}
+                {{- '", "arguments": ' }}
+                {%- if tool_call.arguments is string %}
+                    {{- tool_call.arguments }}
+                {%- else %}
+                    {{- tool_call.arguments | tojson }}
+                {%- endif %}
+                {{- '}\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- endif %}
+{%- endif %}

config.json

@@ -0,0 +1,61 @@
+{
+  "architectures": [
+    "Qwen3ForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "bos_token_id": 151643,
+  "dtype": "float32",
+  "eos_token_id": 151645,
+  "head_dim": 128,
+  "hidden_act": "silu",
+  "hidden_size": 1024,
+  "initializer_range": 0.02,
+  "intermediate_size": 3072,
+  "layer_types": [
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention"
+  ],
+  "max_position_embeddings": 40960,
+  "max_window_layers": 28,
+  "model_type": "qwen3",
+  "num_attention_heads": 16,
+  "num_hidden_layers": 28,
+  "num_key_value_heads": 8,
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": null,
+  "rope_theta": 1000000,
+  "sliding_window": null,
+  "tie_word_embeddings": true,
+  "transformers_version": "4.56.1",
+  "use_cache": false,
+  "use_grouped_mm": true,
+  "use_sliding_window": false,
+  "vocab_size": 151936
+}

generation_config.json

@@ -0,0 +1,13 @@
+{
+  "bos_token_id": 151643,
+  "do_sample": true,
+  "eos_token_id": [
+    151645,
+    151643
+  ],
+  "pad_token_id": 151643,
+  "temperature": 0.6,
+  "top_k": 20,
+  "top_p": 0.95,
+  "transformers_version": "4.56.1"
+}

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a0a127ed2d3885284d216e87de6ad47cb30d155ff020976c1b315030b02cb001
+size 1503365567

special_tokens_map.json

@@ -0,0 +1,25 @@
+{
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "eos_token": {
+    "content": "<|im_end|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": "<|im_end|>"
+}

tokenizer_config.json

@@ -0,0 +1,239 @@
+{
+  "add_bos_token": false,
+  "add_prefix_space": false,
+  "added_tokens_decoder": {
+    "151643": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151644": {
+      "content": "<|im_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151645": {
+      "content": "<|im_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151646": {
+      "content": "<|object_ref_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151647": {
+      "content": "<|object_ref_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151648": {
+      "content": "<|box_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151649": {
+      "content": "<|box_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151650": {
+      "content": "<|quad_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151651": {
+      "content": "<|quad_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151652": {
+      "content": "<|vision_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151653": {
+      "content": "<|vision_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151654": {
+      "content": "<|vision_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151655": {
+      "content": "<|image_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151656": {
+      "content": "<|video_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151657": {
+      "content": "<tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151658": {
+      "content": "</tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151659": {
+      "content": "<|fim_prefix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151660": {
+      "content": "<|fim_middle|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151661": {
+      "content": "<|fim_suffix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151662": {
+      "content": "<|fim_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151663": {
+      "content": "<|repo_name|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151664": {
+      "content": "<|file_sep|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151665": {
+      "content": "<tool_response>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151666": {
+      "content": "</tool_response>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151667": {
+      "content": "<think>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151668": {
+      "content": "</think>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    }
+  },
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "extra_special_tokens": {},
+  "model_max_length": 131072,
+  "pad_token": "<|im_end|>",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null
+}