Hugging Face's logo

cerebras
/
GLM-4.5-Air-REAP-82B-A12B-FP8

Text Generation
Transformers
Safetensors
English
glm4_moe
glm
MOE
pruning
compression
conversational
compressed-tensors
Model card Files
xet
Community
lazarevich commited on
Commit
50252f1
·
0 Parent(s):

upload checkpoint

Browse files
Files changed (25) hide show
  1. README.md +137 -0
  2. chat_template.jinja +103 -0
  3. config.json +420 -0
  4. generation_config.json +10 -0
  5. model-00001-of-00017.safetensors +3 -0
  6. model-00002-of-00017.safetensors +3 -0
  7. model-00003-of-00017.safetensors +3 -0
  8. model-00004-of-00017.safetensors +3 -0
  9. model-00005-of-00017.safetensors +3 -0
  10. model-00006-of-00017.safetensors +3 -0
  11. model-00007-of-00017.safetensors +3 -0
  12. model-00008-of-00017.safetensors +3 -0
  13. model-00009-of-00017.safetensors +3 -0
  14. model-00010-of-00017.safetensors +3 -0
  15. model-00011-of-00017.safetensors +3 -0
  16. model-00012-of-00017.safetensors +3 -0
  17. model-00013-of-00017.safetensors +3 -0
  18. model-00014-of-00017.safetensors +3 -0
  19. model-00015-of-00017.safetensors +3 -0
  20. model-00016-of-00017.safetensors +3 -0
  21. model-00017-of-00017.safetensors +3 -0
  22. model.safetensors.index.json +0 -0
  23. special_tokens_map.json +40 -0
  24. tokenizer.json +3 -0
  25. tokenizer_config.json +325 -0
README.md ADDED
@@ -0,0 +1,137 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language:
3
+ - en
4
+ library_name: transformers
5
+ tags:
6
+ - glm
7
+ - MOE
8
+ - pruning
9
+ - compression
10
+ license: mit
11
+ name: cerebras/GLM-4.5-Air-REAP-82B-A12B-FP8
12
+ description: >
13
+ This model was obtained by uniformly pruning 25% of experts in GLM-4.5-Air-FP8 using the REAP method.
14
+ readme: >
15
+ https://huggingface.co/cerebras/GLM-4.5-Air-REAP-82B-A12B-FP8/main/README.md
16
+ license_link: https://huggingface.co/zai-org/GLM-4.5-Air-FP8/blob/main/LICENSE
17
+ pipeline_tag: text-generation
18
+ base_model:
19
+ - zai-org/GLM-4.5-Air-FP8
20
+ ---
21
+
22
+ <p align="center">
23
+ <em>𓌳 <strong>REAP</strong>𓌳 the Experts: Why Pruning Prevails for One-Shot MoE Compression</em><br>
24
+ <img src="https://i.imgur.com/rmzG3gg.png" alt="REAP" width="75%">
25
+ </p>
26
+
27
+ # GLM-4.5-Air-REAP-82B-A12B-FP8
28
+
29
+ ## ✨ Highlights
30
+
31
+ Introducing **GLM-4.5-Air-REAP-82B-A12B-FP8**, a **memory-efficient compressed variant** of GLM-4.5-Air-FP8 that maintains near-identical performance while being **25% lighter**.
32
+
33
+ This model was created using **REAP (Router-weighted Expert Activation Pruning)**, a novel expert pruning method that selectively removes redundant experts while preserving the router's independent control over remaining experts. Key features include:
34
+
35
+ - **Near-Lossless Performance**: Maintains almost identical accuracy on code generation, agentic coding, and function calling tasks compared to the full 106B model
36
+ - **25% Memory Reduction**: Compressed from 106B to 82B parameters, significantly lowering deployment costs and memory requirements
37
+ - **Preserved Capabilities**: Retains all core functionalities including code generation, agentic workflows, repository-scale understanding, and function calling
38
+ - **Drop-in Compatibility**: Works with vanilla vLLM - no source modifications or custom patches required
39
+ - **Optimized for Real-World Use**: Particularly effective for resource-constrained environments, local deployments, and academic research
40
+
41
+ Note: a BF16 version for more accurate downstream low-bit quantization [is also available on HF](https://huggingface.co/cerebras/GLM-4.5-Air-REAP-82B-A12B).
42
+
43
+ ---
44
+ ## 📋 Model Overview
45
+
46
+ **GLM-4.5-Air-REAP-82B-A12B-FP8** has the following specifications:
47
+
48
+ - **Base Model**: GLM-4.5-Air-FP8
49
+ - **Compression Method**: REAP (Router-weighted Expert Activation Pruning)
50
+ - **Compression Ratio**: 25% expert pruning
51
+ - **Type**: Sparse Mixture-of-Experts (SMoE) Causal Language Model
52
+ - **Number of Parameters**: 82B total, 12B activated per token
53
+ - **Number of Layers**: 46
54
+ - **Number of Attention Heads (GQA)**: 96 for Q and 8 for KV
55
+ - **Number of Experts**: 96 (uniformly pruned from 128)
56
+ - **Number of Activated Experts**: 8 per token
57
+ - **Context Length**: 131,072 tokens
58
+ - **License**: MIT
59
+
60
+ ---
61
+
62
+ ## 📊 Evaluations
63
+
64
+ TBD for FP8 model. Evalulation results [available for the BF16 variant](https://huggingface.co/cerebras/GLM-4.5-Air-REAP-82B-A12B#%F0%9F%93%8A-evaluations).
65
+
66
+ For more details on the evaluation setup, refer to the [REAP arXiv preprint](https://arxiv.org/abs/2510.13999).
67
+
68
+ ---
69
+
70
+ ## 🚀 Deployment
71
+
72
+ You can deploy the model directly using the **latest vLLM** (v0.11.0), no source modifications or custom patches required.
73
+
74
+ ```bash
75
+ vllm serve cerebras/GLM-4.5-Air-REAP-82B-A12B-FP8 \
76
+ --tensor-parallel-size 4 \
77
+ --tool-call-parser glm45 \
78
+ --enable-auto-tool-choice \
79
+ --enable-expert-parallel
80
+ ```
81
+
82
+ If you encounter insufficient memory when running this model, you might need to set a lower value for `--max-num-seqs` flag (e.g. set to 64).
83
+
84
+
85
+ ## 🧩 Model Creation
86
+
87
+ This checkpoint was created by applying the **REAP (Router-weighted Expert Activation Pruning)** method uniformly across all Mixture-of-Experts (MoE) blocks of **GLM-4.5-Air-FP8**, with a **25% pruning rate**.
88
+
89
+ ### How REAP Works
90
+
91
+ REAP selects experts to prune based on a novel **saliency criterion** that considers both:
92
+ - **Router gate values**: How frequently and strongly the router activates each expert
93
+ - **Expert activation norms**: The magnitude of each expert's output contributions
94
+
95
+ This dual consideration ensures that experts contributing minimally to the layer's output are pruned, while preserving those that play critical roles in the model's computations.
96
+
97
+ ### Key Advantages
98
+
99
+ - **One-Shot Compression**: No fine-tuning required after pruning - the model is immediately ready for deployment
100
+ - **Preserved Router Control**: Unlike expert merging methods, REAP maintains the router's independent, input-dependent control over remaining experts, avoiding "functional subspace collapse"
101
+ - **Generative Task Superiority**: REAP significantly outperforms expert merging approaches on generative benchmarks (code generation, creative writing, mathematical reasoning) while maintaining competitive performance on discriminative tasks
102
+
103
+ ### Calibration
104
+
105
+ The model was calibrated using a diverse mixture of domain-specific datasets including:
106
+ - Code generation samples ([evol-codealpaca](https://huggingface.co/datasets/theblackcat102/evol-codealpaca-v1))
107
+ - Function calling examples ([xlam-function-calling](https://huggingface.co/datasets/Salesforce/xlam-function-calling-60k))
108
+ - Agentic multi-turn trajectories ([SWE-smith-trajectories](https://huggingface.co/datasets/SWE-bench/SWE-smith-trajectories))
109
+
110
+ 📚 For more details, refer to the following resources:
111
+
112
+ - [🧾 arXiv Preprint](https://arxiv.org/abs/2510.13999)
113
+ - [🧾 REAP Blog](https://www.cerebras.ai/blog/reap)
114
+ - [💻 REAP Codebase (GitHub)](https://github.com/CerebrasResearch/reap)
115
+
116
+ ---
117
+
118
+ ## ⚖️ License
119
+
120
+ This model is derived from
121
+ **[`zai-org/GLM-4.5-Air-FP8`](https://huggingface.co/zai-org/GLM-4.5-Air-FP8)**
122
+ and distributed under the **MIT license**.
123
+
124
+ ---
125
+
126
+ ## 🧾 Citation
127
+
128
+ If you use this checkpoint, please cite the REAP paper:
129
+
130
+ ```bibtex
131
+ @article{lasby-reap,
132
+ title={REAP the Experts: Why Pruning Prevails for One-Shot MoE compression},
133
+ author={Lasby, Mike and Lazarevich, Ivan and Sinnadurai, Nish and Lie, Sean and Ioannou, Yani and Thangarasa, Vithursan},
134
+ journal={arXiv preprint arXiv:2510.13999},
135
+ year={2025}
136
+ }
137
+ ```
chat_template.jinja ADDED
@@ -0,0 +1,103 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [gMASK]<sop>
2
+ {%- if tools -%}
3
+ <|system|>
4
+ # Tools
5
+
6
+ You may call one or more functions to assist with the user query.
7
+
8
+ You are provided with function signatures within <tools></tools> XML tags:
9
+ <tools>
10
+ {% for tool in tools %}
11
+ {{ tool | tojson(ensure_ascii=False) }}
12
+ {% endfor %}
13
+ </tools>
14
+
15
+ For each function call, output the function name and arguments within the following XML format:
16
+ <tool_call>{function-name}
17
+ <arg_key>{arg-key-1}</arg_key>
18
+ <arg_value>{arg-value-1}</arg_value>
19
+ <arg_key>{arg-key-2}</arg_key>
20
+ <arg_value>{arg-value-2}</arg_value>
21
+ ...
22
+ </tool_call>{%- endif -%}
23
+ {%- macro visible_text(content) -%}
24
+ {%- if content is string -%}
25
+ {{- content }}
26
+ {%- elif content is iterable and content is not mapping -%}
27
+ {%- for item in content -%}
28
+ {%- if item is mapping and item.type == 'text' -%}
29
+ {{- item.text }}
30
+ {%- elif item is string -%}
31
+ {{- item }}
32
+ {%- endif -%}
33
+ {%- endfor -%}
34
+ {%- else -%}
35
+ {{- content }}
36
+ {%- endif -%}
37
+ {%- endmacro -%}
38
+ {%- set ns = namespace(last_user_index=-1) %}
39
+ {%- for m in messages %}
40
+ {%- if m.role == 'user' %}
41
+ {% set ns.last_user_index = loop.index0 -%}
42
+ {%- endif %}
43
+ {%- endfor %}
44
+ {% for m in messages %}
45
+ {%- if m.role == 'user' -%}<|user|>
46
+ {{ visible_text(m.content) }}
47
+ {{- '/nothink' if (enable_thinking is defined and not enable_thinking and not visible_text(m.content).endswith("/nothink")) else '' -}}
48
+ {%- elif m.role == 'assistant' -%}
49
+ <|assistant|>
50
+ {%- set reasoning_content = '' %}
51
+ {%- set content = visible_text(m.content) %}
52
+ {%- if m.reasoning_content is string %}
53
+ {%- set reasoning_content = m.reasoning_content %}
54
+ {%- else %}
55
+ {%- if '</think>' in content %}
56
+ {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
57
+ {%- set content = content.split('</think>')[-1].lstrip('\n') %}
58
+ {%- endif %}
59
+ {%- endif %}
60
+ {%- if loop.index0 > ns.last_user_index and reasoning_content -%}
61
+ {{ '\n<think>' + reasoning_content.strip() + '</think>'}}
62
+ {%- else -%}
63
+ {{ '\n<think></think>' }}
64
+ {%- endif -%}
65
+ {%- if content.strip() -%}
66
+ {{ '\n' + content.strip() }}
67
+ {%- endif -%}
68
+ {% if m.tool_calls %}
69
+ {% for tc in m.tool_calls %}
70
+ {%- if tc.function %}
71
+ {%- set tc = tc.function %}
72
+ {%- endif %}
73
+ {{ '\n<tool_call>' + tc.name }}
74
+ {% set _args = tc.arguments %}
75
+ {% for k, v in _args.items() %}
76
+ <arg_key>{{ k }}</arg_key>
77
+ <arg_value>{{ v | tojson(ensure_ascii=False) if v is not string else v }}</arg_value>
78
+ {% endfor %}
79
+ </tool_call>{% endfor %}
80
+ {% endif %}
81
+ {%- elif m.role == 'tool' -%}
82
+ {%- if m.content is string -%}
83
+ {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
84
+ {{- '<|observation|>' }}
85
+ {%- endif %}
86
+ {{- '\n<tool_response>\n' }}
87
+ {{- m.content }}
88
+ {{- '\n</tool_response>' }}
89
+ {%- else -%}
90
+ <|observation|>{% for tr in m.content %}
91
+
92
+ <tool_response>
93
+ {{ tr.output if tr.output is defined else tr }}
94
+ </tool_response>{% endfor -%}
95
+ {% endif -%}
96
+ {%- elif m.role == 'system' -%}
97
+ <|system|>
98
+ {{ visible_text(m.content) }}
99
+ {%- endif -%}
100
+ {%- endfor -%}
101
+ {%- if add_generation_prompt -%}
102
+ <|assistant|>{{- '\n<think></think>' if (enable_thinking is defined and not enable_thinking) else '' -}}
103
+ {%- endif -%}
config.json ADDED
@@ -0,0 +1,420 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Glm4MoeForCausalLM"
4
+ ],
5
+ "attention_bias": true,
6
+ "attention_dropout": 0.0,
7
+ "eos_token_id": [
8
+ 151329,
9
+ 151336,
10
+ 151338
11
+ ],
12
+ "first_k_dense_replace": 1,
13
+ "head_dim": 128,
14
+ "hidden_act": "silu",
15
+ "hidden_size": 4096,
16
+ "initializer_range": 0.02,
17
+ "intermediate_size": 10944,
18
+ "max_position_embeddings": 131072,
19
+ "model_type": "glm4_moe",
20
+ "moe_intermediate_size": 1408,
21
+ "n_group": 1,
22
+ "n_routed_experts": 96,
23
+ "n_shared_experts": 1,
24
+ "norm_topk_prob": true,
25
+ "num_attention_heads": 96,
26
+ "num_experts_per_tok": 8,
27
+ "num_hidden_layers": 46,
28
+ "num_key_value_heads": 8,
29
+ "num_nextn_predict_layers": 0,
30
+ "pad_token_id": 151329,
31
+ "partial_rotary_factor": 0.5,
32
+ "quantization_config": {
33
+ "config_groups": {
34
+ "group_0": {
35
+ "input_activations": {
36
+ "actorder": null,
37
+ "block_structure": null,
38
+ "dynamic": true,
39
+ "group_size": null,
40
+ "num_bits": 8,
41
+ "observer": null,
42
+ "observer_kwargs": {},
43
+ "strategy": "token",
44
+ "symmetric": true,
45
+ "type": "float"
46
+ },
47
+ "output_activations": null,
48
+ "targets": [
49
+ "Linear"
50
+ ],
51
+ "weights": {
52
+ "actorder": null,
53
+ "block_structure": null,
54
+ "dynamic": false,
55
+ "group_size": null,
56
+ "num_bits": 8,
57
+ "observer": "minmax",
58
+ "observer_kwargs": {},
59
+ "strategy": "channel",
60
+ "symmetric": true,
61
+ "type": "float"
62
+ }
63
+ }
64
+ },
65
+ "format": "float-quantized",
66
+ "global_compression_ratio": null,
67
+ "ignore": [
68
+ "model.layers.9.mlp.gate",
69
+ "model.layers.22.self_attn.q_proj.bias",
70
+ "model.layers.21.post_attention_layernorm",
71
+ "model.layers.46.post_attention_layernorm",
72
+ "model.layers.33.mlp.gate.e_score_correction_bias",
73
+ "model.layers.19.input_layernorm",
74
+ "model.layers.23.input_layernorm",
75
+ "model.layers.26.input_layernorm",
76
+ "model.layers.16.input_layernorm",
77
+ "model.layers.2.input_layernorm",
78
+ "model.layers.43.mlp.gate.e_score_correction_bias",
79
+ "model.layers.17.self_attn.v_proj.bias",
80
+ "model.layers.4.self_attn.k_proj.bias",
81
+ "model.layers.24.mlp.gate",
82
+ "model.layers.42.self_attn.q_proj.bias",
83
+ "model.layers.19.self_attn.v_proj.bias",
84
+ "model.layers.36.self_attn.k_proj.bias",
85
+ "model.layers.43.self_attn.k_proj.bias",
86
+ "model.layers.26.self_attn.v_proj.bias",
87
+ "model.layers.33.mlp.gate",
88
+ "model.layers.31.input_layernorm",
89
+ "model.layers.8.input_layernorm",
90
+ "model.layers.9.mlp.gate.e_score_correction_bias",
91
+ "model.layers.19.mlp.gate.e_score_correction_bias",
92
+ "model.layers.14.self_attn.v_proj.bias",
93
+ "model.layers.19.mlp.gate",
94
+ "model.layers.12.mlp.gate",
95
+ "model.layers.37.mlp.gate.e_score_correction_bias",
96
+ "model.layers.39.mlp.gate",
97
+ "model.layers.20.self_attn.v_proj.bias",
98
+ "model.layers.1.self_attn.q_proj.bias",
99
+ "model.layers.7.self_attn.k_proj.bias",
100
+ "model.layers.19.self_attn.k_proj.bias",
101
+ "model.layers.30.mlp.gate.e_score_correction_bias",
102
+ "model.layers.37.input_layernorm",
103
+ "model.layers.40.self_attn.q_proj.bias",
104
+ "model.layers.30.input_layernorm",
105
+ "model.layers.10.post_attention_layernorm",
106
+ "model.layers.10.mlp.gate",
107
+ "model.layers.21.mlp.gate",
108
+ "model.layers.39.post_attention_layernorm",
109
+ "model.layers.20.self_attn.q_proj.bias",
110
+ "model.layers.20.input_layernorm",
111
+ "model.layers.16.self_attn.v_proj.bias",
112
+ "model.layers.5.self_attn.k_proj.bias",
113
+ "model.layers.18.input_layernorm",
114
+ "model.layers.35.input_layernorm",
115
+ "model.layers.43.mlp.gate",
116
+ "model.layers.5.self_attn.v_proj.bias",
117
+ "model.layers.28.mlp.gate",
118
+ "model.layers.4.self_attn.v_proj.bias",
119
+ "model.layers.46.mlp.gate.e_score_correction_bias",
120
+ "model.layers.25.mlp.gate.e_score_correction_bias",
121
+ "model.layers.46.enorm",
122
+ "model.layers.30.self_attn.k_proj.bias",
123
+ "model.layers.27.self_attn.v_proj.bias",
124
+ "model.layers.23.mlp.gate",
125
+ "model.layers.38.post_attention_layernorm",
126
+ "model.layers.0.post_attention_layernorm",
127
+ "model.layers.46.self_attn.k_proj.bias",
128
+ "model.layers.26.post_attention_layernorm",
129
+ "model.layers.26.self_attn.q_proj.bias",
130
+ "model.layers.46.embed_tokens",
131
+ "model.layers.24.input_layernorm",
132
+ "model.layers.41.post_attention_layernorm",
133
+ "model.layers.20.mlp.gate",
134
+ "model.layers.3.mlp.gate",
135
+ "model.layers.22.input_layernorm",
136
+ "model.layers.15.post_attention_layernorm",
137
+ "model.layers.39.input_layernorm",
138
+ "model.layers.42.mlp.gate.e_score_correction_bias",
139
+ "model.layers.34.mlp.gate",
140
+ "model.layers.13.mlp.gate",
141
+ "model.layers.38.input_layernorm",
142
+ "model.layers.15.self_attn.q_proj.bias",
143
+ "model.layers.7.post_attention_layernorm",
144
+ "model.layers.28.self_attn.v_proj.bias",
145
+ "model.layers.36.post_attention_layernorm",
146
+ "model.layers.34.self_attn.k_proj.bias",
147
+ "model.layers.23.post_attention_layernorm",
148
+ "model.layers.43.input_layernorm",
149
+ "model.layers.39.self_attn.k_proj.bias",
150
+ "model.layers.18.post_attention_layernorm",
151
+ "model.layers.29.mlp.gate.e_score_correction_bias",
152
+ "model.layers.34.mlp.gate.e_score_correction_bias",
153
+ "model.layers.28.self_attn.q_proj.bias",
154
+ "model.layers.46.eh_proj",
155
+ "model.layers.14.mlp.gate.e_score_correction_bias",
156
+ "model.layers.7.mlp.gate.e_score_correction_bias",
157
+ "model.layers.35.self_attn.k_proj.bias",
158
+ "model.layers.13.post_attention_layernorm",
159
+ "model.layers.22.self_attn.k_proj.bias",
160
+ "model.layers.4.self_attn.q_proj.bias",
161
+ "model.layers.41.self_attn.k_proj.bias",
162
+ "model.layers.12.post_attention_layernorm",
163
+ "model.layers.37.self_attn.q_proj.bias",
164
+ "model.layers.46.input_layernorm",
165
+ "model.layers.24.self_attn.k_proj.bias",
166
+ "model.layers.5.mlp.gate",
167
+ "model.layers.9.self_attn.k_proj.bias",
168
+ "model.layers.10.self_attn.v_proj.bias",
169
+ "model.layers.42.self_attn.v_proj.bias",
170
+ "model.embed_tokens",
171
+ "model.layers.2.self_attn.q_proj.bias",
172
+ "model.layers.28.mlp.gate.e_score_correction_bias",
173
+ "model.layers.24.self_attn.v_proj.bias",
174
+ "model.layers.15.input_layernorm",
175
+ "model.layers.9.input_layernorm",
176
+ "model.layers.33.input_layernorm",
177
+ "model.layers.45.self_attn.v_proj.bias",
178
+ "model.layers.31.self_attn.q_proj.bias",
179
+ "model.layers.34.input_layernorm",
180
+ "model.layers.14.input_layernorm",
181
+ "model.layers.17.post_attention_layernorm",
182
+ "model.layers.0.self_attn.k_proj.bias",
183
+ "model.layers.37.self_attn.v_proj.bias",
184
+ "model.norm",
185
+ "model.layers.9.self_attn.q_proj.bias",
186
+ "model.layers.4.input_layernorm",
187
+ "model.layers.45.self_attn.q_proj.bias",
188
+ "model.layers.7.self_attn.q_proj.bias",
189
+ "model.layers.32.self_attn.v_proj.bias",
190
+ "model.layers.22.self_attn.v_proj.bias",
191
+ "model.layers.45.post_attention_layernorm",
192
+ "model.layers.40.mlp.gate",
193
+ "model.layers.29.self_attn.v_proj.bias",
194
+ "model.layers.3.mlp.gate.e_score_correction_bias",
195
+ "model.layers.31.post_attention_layernorm",
196
+ "model.layers.41.self_attn.v_proj.bias",
197
+ "model.layers.5.input_layernorm",
198
+ "model.layers.13.self_attn.v_proj.bias",
199
+ "model.layers.26.self_attn.k_proj.bias",
200
+ "model.layers.28.post_attention_layernorm",
201
+ "model.layers.17.mlp.gate",
202
+ "model.layers.42.mlp.gate",
203
+ "model.layers.34.self_attn.v_proj.bias",
204
+ "model.layers.1.mlp.gate.e_score_correction_bias",
205
+ "model.layers.21.input_layernorm",
206
+ "model.layers.21.self_attn.k_proj.bias",
207
+ "model.layers.29.self_attn.k_proj.bias",
208
+ "model.layers.20.post_attention_layernorm",
209
+ "model.layers.14.post_attention_layernorm",
210
+ "model.layers.34.post_attention_layernorm",
211
+ "model.layers.27.self_attn.k_proj.bias",
212
+ "model.layers.24.mlp.gate.e_score_correction_bias",
213
+ "model.layers.31.mlp.gate.e_score_correction_bias",
214
+ "model.layers.2.self_attn.k_proj.bias",
215
+ "model.layers.25.self_attn.v_proj.bias",
216
+ "model.layers.1.post_attention_layernorm",
217
+ "model.layers.10.self_attn.q_proj.bias",
218
+ "model.layers.16.mlp.gate.e_score_correction_bias",
219
+ "model.layers.16.self_attn.q_proj.bias",
220
+ "model.layers.38.mlp.gate.e_score_correction_bias",
221
+ "model.layers.46.self_attn.q_proj.bias",
222
+ "model.layers.23.self_attn.k_proj.bias",
223
+ "model.layers.42.post_attention_layernorm",
224
+ "model.layers.33.self_attn.k_proj.bias",
225
+ "model.layers.30.mlp.gate",
226
+ "model.layers.34.self_attn.q_proj.bias",
227
+ "model.layers.4.post_attention_layernorm",
228
+ "model.layers.13.self_attn.k_proj.bias",
229
+ "model.layers.2.post_attention_layernorm",
230
+ "model.layers.40.post_attention_layernorm",
231
+ "model.layers.38.self_attn.k_proj.bias",
232
+ "model.layers.1.self_attn.k_proj.bias",
233
+ "model.layers.10.mlp.gate.e_score_correction_bias",
234
+ "model.layers.43.self_attn.v_proj.bias",
235
+ "model.layers.11.input_layernorm",
236
+ "model.layers.42.input_layernorm",
237
+ "model.layers.19.self_attn.q_proj.bias",
238
+ "model.layers.24.post_attention_layernorm",
239
+ "model.layers.12.input_layernorm",
240
+ "model.layers.42.self_attn.k_proj.bias",
241
+ "model.layers.12.self_attn.k_proj.bias",
242
+ "model.layers.0.self_attn.v_proj.bias",
243
+ "model.layers.1.mlp.gate",
244
+ "model.layers.39.self_attn.v_proj.bias",
245
+ "model.layers.14.mlp.gate",
246
+ "model.layers.44.post_attention_layernorm",
247
+ "model.layers.37.mlp.gate",
248
+ "model.layers.31.mlp.gate",
249
+ "model.layers.8.post_attention_layernorm",
250
+ "model.layers.2.mlp.gate.e_score_correction_bias",
251
+ "model.layers.36.input_layernorm",
252
+ "model.layers.30.post_attention_layernorm",
253
+ "model.layers.46.shared_head.norm",
254
+ "model.layers.4.mlp.gate",
255
+ "model.layers.6.mlp.gate",
256
+ "model.layers.29.mlp.gate",
257
+ "model.layers.7.mlp.gate",
258
+ "model.layers.0.self_attn.q_proj.bias",
259
+ "model.layers.44.mlp.gate",
260
+ "model.layers.32.self_attn.k_proj.bias",
261
+ "model.layers.4.mlp.gate.e_score_correction_bias",
262
+ "model.layers.18.self_attn.v_proj.bias",
263
+ "model.layers.30.self_attn.q_proj.bias",
264
+ "model.layers.21.mlp.gate.e_score_correction_bias",
265
+ "model.layers.32.post_attention_layernorm",
266
+ "model.layers.19.post_attention_layernorm",
267
+ "model.layers.22.mlp.gate",
268
+ "model.layers.13.mlp.gate.e_score_correction_bias",
269
+ "model.layers.8.mlp.gate",
270
+ "model.layers.36.self_attn.v_proj.bias",
271
+ "model.layers.5.post_attention_layernorm",
272
+ "model.layers.32.input_layernorm",
273
+ "model.layers.33.post_attention_layernorm",
274
+ "model.layers.21.self_attn.v_proj.bias",
275
+ "model.layers.2.mlp.gate",
276
+ "model.layers.13.input_layernorm",
277
+ "model.layers.15.self_attn.v_proj.bias",
278
+ "model.layers.16.self_attn.k_proj.bias",
279
+ "model.layers.2.self_attn.v_proj.bias",
280
+ "model.layers.43.post_attention_layernorm",
281
+ "model.layers.7.input_layernorm",
282
+ "model.layers.29.post_attention_layernorm",
283
+ "model.layers.20.self_attn.k_proj.bias",
284
+ "model.layers.38.mlp.gate",
285
+ "model.layers.18.mlp.gate.e_score_correction_bias",
286
+ "model.layers.25.input_layernorm",
287
+ "model.layers.1.input_layernorm",
288
+ "model.layers.46.hnorm",
289
+ "model.layers.31.self_attn.v_proj.bias",
290
+ "model.layers.14.self_attn.q_proj.bias",
291
+ "model.layers.18.self_attn.q_proj.bias",
292
+ "model.layers.8.self_attn.q_proj.bias",
293
+ "model.layers.35.self_attn.v_proj.bias",
294
+ "model.layers.45.mlp.gate.e_score_correction_bias",
295
+ "model.layers.9.post_attention_layernorm",
296
+ "model.layers.30.self_attn.v_proj.bias",
297
+ "model.layers.15.mlp.gate",
298
+ "model.layers.10.input_layernorm",
299
+ "model.layers.6.self_attn.q_proj.bias",
300
+ "model.layers.11.mlp.gate.e_score_correction_bias",
301
+ "model.layers.41.input_layernorm",
302
+ "model.layers.22.mlp.gate.e_score_correction_bias",
303
+ "model.layers.15.mlp.gate.e_score_correction_bias",
304
+ "model.layers.21.self_attn.q_proj.bias",
305
+ "model.layers.17.mlp.gate.e_score_correction_bias",
306
+ "model.layers.16.mlp.gate",
307
+ "model.layers.25.self_attn.q_proj.bias",
308
+ "model.layers.6.input_layernorm",
309
+ "model.layers.17.input_layernorm",
310
+ "model.layers.26.mlp.gate.e_score_correction_bias",
311
+ "model.layers.35.mlp.gate.e_score_correction_bias",
312
+ "model.layers.0.input_layernorm",
313
+ "model.layers.3.post_attention_layernorm",
314
+ "model.layers.6.self_attn.v_proj.bias",
315
+ "model.layers.27.mlp.gate.e_score_correction_bias",
316
+ "model.layers.18.mlp.gate",
317
+ "model.layers.28.input_layernorm",
318
+ "model.layers.9.self_attn.v_proj.bias",
319
+ "model.layers.31.self_attn.k_proj.bias",
320
+ "model.layers.40.self_attn.v_proj.bias",
321
+ "model.layers.12.self_attn.q_proj.bias",
322
+ "model.layers.41.mlp.gate",
323
+ "model.layers.5.self_attn.q_proj.bias",
324
+ "model.layers.11.self_attn.v_proj.bias",
325
+ "model.layers.36.mlp.gate",
326
+ "model.layers.27.self_attn.q_proj.bias",
327
+ "model.layers.40.self_attn.k_proj.bias",
328
+ "model.layers.11.post_attention_layernorm",
329
+ "model.layers.27.input_layernorm",
330
+ "model.layers.12.self_attn.v_proj.bias",
331
+ "model.layers.46.mlp.gate",
332
+ "model.layers.17.self_attn.k_proj.bias",
333
+ "model.layers.3.input_layernorm",
334
+ "model.layers.44.input_layernorm",
335
+ "model.layers.10.self_attn.k_proj.bias",
336
+ "model.layers.41.mlp.gate.e_score_correction_bias",
337
+ "model.layers.7.self_attn.v_proj.bias",
338
+ "model.layers.18.self_attn.k_proj.bias",
339
+ "model.layers.1.self_attn.v_proj.bias",
340
+ "model.layers.26.mlp.gate",
341
+ "model.layers.45.input_layernorm",
342
+ "model.layers.23.self_attn.v_proj.bias",
343
+ "model.layers.39.mlp.gate.e_score_correction_bias",
344
+ "model.layers.12.mlp.gate.e_score_correction_bias",
345
+ "model.layers.37.post_attention_layernorm",
346
+ "model.layers.46.self_attn.v_proj.bias",
347
+ "model.layers.36.mlp.gate.e_score_correction_bias",
348
+ "model.layers.5.mlp.gate.e_score_correction_bias",
349
+ "model.layers.35.mlp.gate",
350
+ "model.layers.44.self_attn.k_proj.bias",
351
+ "model.layers.3.self_attn.k_proj.bias",
352
+ "model.layers.11.mlp.gate",
353
+ "model.layers.11.self_attn.q_proj.bias",
354
+ "model.layers.17.self_attn.q_proj.bias",
355
+ "model.layers.32.self_attn.q_proj.bias",
356
+ "model.layers.11.self_attn.k_proj.bias",
357
+ "model.layers.40.mlp.gate.e_score_correction_bias",
358
+ "model.layers.41.self_attn.q_proj.bias",
359
+ "model.layers.15.self_attn.k_proj.bias",
360
+ "model.layers.44.self_attn.v_proj.bias",
361
+ "model.layers.25.self_attn.k_proj.bias",
362
+ "model.layers.25.post_attention_layernorm",
363
+ "model.layers.29.input_layernorm",
364
+ "model.layers.44.self_attn.q_proj.bias",
365
+ "model.layers.16.post_attention_layernorm",
366
+ "model.layers.6.mlp.gate.e_score_correction_bias",
367
+ "model.layers.38.self_attn.v_proj.bias",
368
+ "model.layers.40.input_layernorm",
369
+ "model.layers.6.post_attention_layernorm",
370
+ "model.layers.22.post_attention_layernorm",
371
+ "model.layers.8.self_attn.k_proj.bias",
372
+ "model.layers.37.self_attn.k_proj.bias",
373
+ "model.layers.23.mlp.gate.e_score_correction_bias",
374
+ "model.layers.27.mlp.gate",
375
+ "model.layers.8.mlp.gate.e_score_correction_bias",
376
+ "model.layers.28.self_attn.k_proj.bias",
377
+ "model.layers.24.self_attn.q_proj.bias",
378
+ "model.layers.39.self_attn.q_proj.bias",
379
+ "model.layers.36.self_attn.q_proj.bias",
380
+ "model.layers.45.self_attn.k_proj.bias",
381
+ "model.layers.32.mlp.gate.e_score_correction_bias",
382
+ "model.layers.35.self_attn.q_proj.bias",
383
+ "model.layers.33.self_attn.q_proj.bias",
384
+ "model.layers.14.self_attn.k_proj.bias",
385
+ "lm_head",
386
+ "model.layers.3.self_attn.v_proj.bias",
387
+ "model.layers.44.mlp.gate.e_score_correction_bias",
388
+ "model.layers.45.mlp.gate",
389
+ "model.layers.32.mlp.gate",
390
+ "model.layers.33.self_attn.v_proj.bias",
391
+ "model.layers.29.self_attn.q_proj.bias",
392
+ "model.layers.3.self_attn.q_proj.bias",
393
+ "model.layers.35.post_attention_layernorm",
394
+ "model.layers.6.self_attn.k_proj.bias",
395
+ "model.layers.43.self_attn.q_proj.bias",
396
+ "model.layers.20.mlp.gate.e_score_correction_bias",
397
+ "model.layers.8.self_attn.v_proj.bias",
398
+ "model.layers.13.self_attn.q_proj.bias",
399
+ "model.layers.27.post_attention_layernorm",
400
+ "model.layers.38.self_attn.q_proj.bias",
401
+ "model.layers.25.mlp.gate",
402
+ "model.layers.23.self_attn.q_proj.bias"
403
+ ],
404
+ "kv_cache_scheme": null,
405
+ "quant_method": "compressed-tensors",
406
+ "quantization_status": "compressed",
407
+ "sparsity_config": {}
408
+ },
409
+ "rms_norm_eps": 1e-05,
410
+ "rope_scaling": null,
411
+ "rope_theta": 1000000,
412
+ "routed_scaling_factor": 1.0,
413
+ "tie_word_embeddings": false,
414
+ "topk_group": 1,
415
+ "torch_dtype": "bfloat16",
416
+ "transformers_version": "4.55.0",
417
+ "use_cache": true,
418
+ "use_qk_norm": false,
419
+ "vocab_size": 151552
420
+ }
generation_config.json ADDED
@@ -0,0 +1,10 @@
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "eos_token_id": [
4
+ 151329,
5
+ 151336,
6
+ 151338
7
+ ],
8
+ "pad_token_id": 151329,
9
+ "transformers_version": "4.55.0"
10
+ }
model-00001-of-00017.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ce04e1bd0cfa3d1f21f97072476fe99a267bb7f13e5ed303cff24e2b7e5dfd30
3
+ size 5000053512
model-00002-of-00017.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f4c42f285411bbc7a66841c44023ffe817f80de732e937200c6eed3a61cf93ed
3
+ size 4999335584
model-00003-of-00017.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6b313b956dc978de3e2565577f5035939ef04815feffcc4d0a5fda5d500fd3c3
3
+ size 4999335504
model-00004-of-00017.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e8c1948e8da128f112c5686301261c740f8bba4645432c5b7da46db36ab37cab
3
+ size 4999330632
model-00005-of-00017.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c29c5131eb8c6eaf950995a513337439494b737525d627f486fa77808815deb5
3
+ size 4999336928
model-00006-of-00017.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:de537de369b8faf684be04253123406e8b2d97ca2d907e859df22e8740d1eb5b
3
+ size 4999081480
model-00007-of-00017.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6b4d2ba1a7bae48926d27fc4177b91bc5fdae9f73e18724934db27bbde65ecd6
3
+ size 4999337232
model-00008-of-00017.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4a398cc1918950d053523d7470da5072c79638114505512a360994563971c230
3
+ size 4999331776
model-00009-of-00017.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:76575f216bd15dcb5350ee2ddc6d77b92decd88894236af598ae41ffcb1f1aab
3
+ size 4999337024
model-00010-of-00017.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a83229cded615d05a066da05e850056b8eaad7ec9cfbfbfd4fc4746cf7743c6b
3
+ size 4999336928
model-00011-of-00017.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dfdd28fad393daeca562fbc35763f6a26550900b73175a58fc1f07090103ac66
3
+ size 4999081456
model-00012-of-00017.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:803db97559cbeaa2356b28ce453a0332bc2b5295cbe0be77fb7b3b16adbfa70b
3
+ size 4999331856
model-00013-of-00017.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fee9fa3b898ab63d7cb7fb30c7c06efdddb073fc7119c0931c6a0a0cc6bf0f4f
3
+ size 4999337136
model-00014-of-00017.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:636d8943b172db1932a89543d06cd9d5d9f2afe70931ad1e7b690f23f51c4251
3
+ size 4999337000
model-00015-of-00017.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b51635ed4a114a9a7552ae131b9604c158571e42022c6ef82f9e93739ee4c1ec
3
+ size 4999331560
model-00016-of-00017.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8fdd875de11871f53332e3ffb57801266733cf43607e9aed6f6d70814ef4bd0d
3
+ size 4999081440
model-00017-of-00017.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d52447099426a7dffa35b939c3cf59d10e1c8d60df5fbf92b347bd152b331650
3
+ size 3268561760
model.safetensors.index.json ADDED
The diff for this file is too large to render. See raw diff
 
special_tokens_map.json ADDED
@@ -0,0 +1,40 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "additional_special_tokens": [
3
+ "<|endoftext|>",
4
+ "[MASK]",
5
+ "[gMASK]",
6
+ "[sMASK]",
7
+ "<sop>",
8
+ "<eop>",
9
+ "<|system|>",
10
+ "<|user|>",
11
+ "<|assistant|>",
12
+ "<|observation|>",
13
+ "<|begin_of_image|>",
14
+ "<|end_of_image|>",
15
+ "<|begin_of_video|>",
16
+ "<|end_of_video|>",
17
+ "<|begin_of_audio|>",
18
+ "<|end_of_audio|>",
19
+ "<|begin_of_transcription|>",
20
+ "<|end_of_transcription|>",
21
+ "<|code_prefix|>",
22
+ "<|code_middle|>",
23
+ "<|code_suffix|>",
24
+ "/nothink"
25
+ ],
26
+ "eos_token": {
27
+ "content": "<|endoftext|>",
28
+ "lstrip": false,
29
+ "normalized": false,
30
+ "rstrip": false,
31
+ "single_word": false
32
+ },
33
+ "pad_token": {
34
+ "content": "<|endoftext|>",
35
+ "lstrip": false,
36
+ "normalized": false,
37
+ "rstrip": false,
38
+ "single_word": false
39
+ }
40
+ }
tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bda8e2146c3bb7b7e0fc96dcc4f0aeff041c6c27952e3ace0665663ebff346ba
3
+ size 19970700
tokenizer_config.json ADDED
@@ -0,0 +1,325 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "added_tokens_decoder": {
3
+ "151329": {
4
+ "content": "<|endoftext|>",
5
+ "lstrip": false,
6
+ "normalized": false,
7
+ "rstrip": false,
8
+ "single_word": false,
9
+ "special": true
10
+ },
11
+ "151330": {
12
+ "content": "[MASK]",
13
+ "lstrip": false,
14
+ "normalized": false,
15
+ "rstrip": false,
16
+ "single_word": false,
17
+ "special": true
18
+ },
19
+ "151331": {
20
+ "content": "[gMASK]",
21
+ "lstrip": false,
22
+ "normalized": false,
23
+ "rstrip": false,
24
+ "single_word": false,
25
+ "special": true
26
+ },
27
+ "151332": {
28
+ "content": "[sMASK]",
29
+ "lstrip": false,
30
+ "normalized": false,
31
+ "rstrip": false,
32
+ "single_word": false,
33
+ "special": true
34
+ },
35
+ "151333": {
36
+ "content": "<sop>",
37
+ "lstrip": false,
38
+ "normalized": false,
39
+ "rstrip": false,
40
+ "single_word": false,
41
+ "special": true
42
+ },
43
+ "151334": {
44
+ "content": "<eop>",
45
+ "lstrip": false,
46
+ "normalized": false,
47
+ "rstrip": false,
48
+ "single_word": false,
49
+ "special": true
50
+ },
51
+ "151335": {
52
+ "content": "<|system|>",
53
+ "lstrip": false,
54
+ "normalized": false,
55
+ "rstrip": false,
56
+ "single_word": false,
57
+ "special": true
58
+ },
59
+ "151336": {
60
+ "content": "<|user|>",
61
+ "lstrip": false,
62
+ "normalized": false,
63
+ "rstrip": false,
64
+ "single_word": false,
65
+ "special": true
66
+ },
67
+ "151337": {
68
+ "content": "<|assistant|>",
69
+ "lstrip": false,
70
+ "normalized": false,
71
+ "rstrip": false,
72
+ "single_word": false,
73
+ "special": true
74
+ },
75
+ "151338": {
76
+ "content": "<|observation|>",
77
+ "lstrip": false,
78
+ "normalized": false,
79
+ "rstrip": false,
80
+ "single_word": false,
81
+ "special": true
82
+ },
83
+ "151339": {
84
+ "content": "<|begin_of_image|>",
85
+ "lstrip": false,
86
+ "normalized": false,
87
+ "rstrip": false,
88
+ "single_word": false,
89
+ "special": true
90
+ },
91
+ "151340": {
92
+ "content": "<|end_of_image|>",
93
+ "lstrip": false,
94
+ "normalized": false,
95
+ "rstrip": false,
96
+ "single_word": false,
97
+ "special": true
98
+ },
99
+ "151341": {
100
+ "content": "<|begin_of_video|>",
101
+ "lstrip": false,
102
+ "normalized": false,
103
+ "rstrip": false,
104
+ "single_word": false,
105
+ "special": true
106
+ },
107
+ "151342": {
108
+ "content": "<|end_of_video|>",
109
+ "lstrip": false,
110
+ "normalized": false,
111
+ "rstrip": false,
112
+ "single_word": false,
113
+ "special": true
114
+ },
115
+ "151343": {
116
+ "content": "<|begin_of_audio|>",
117
+ "lstrip": false,
118
+ "normalized": false,
119
+ "rstrip": false,
120
+ "single_word": false,
121
+ "special": true
122
+ },
123
+ "151344": {
124
+ "content": "<|end_of_audio|>",
125
+ "lstrip": false,
126
+ "normalized": false,
127
+ "rstrip": false,
128
+ "single_word": false,
129
+ "special": true
130
+ },
131
+ "151345": {
132
+ "content": "<|begin_of_transcription|>",
133
+ "lstrip": false,
134
+ "normalized": false,
135
+ "rstrip": false,
136
+ "single_word": false,
137
+ "special": true
138
+ },
139
+ "151346": {
140
+ "content": "<|end_of_transcription|>",
141
+ "lstrip": false,
142
+ "normalized": false,
143
+ "rstrip": false,
144
+ "single_word": false,
145
+ "special": true
146
+ },
147
+ "151347": {
148
+ "content": "<|code_prefix|>",
149
+ "lstrip": false,
150
+ "normalized": false,
151
+ "rstrip": false,
152
+ "single_word": false,
153
+ "special": true
154
+ },
155
+ "151348": {
156
+ "content": "<|code_middle|>",
157
+ "lstrip": false,
158
+ "normalized": false,
159
+ "rstrip": false,
160
+ "single_word": false,
161
+ "special": true
162
+ },
163
+ "151349": {
164
+ "content": "<|code_suffix|>",
165
+ "lstrip": false,
166
+ "normalized": false,
167
+ "rstrip": false,
168
+ "single_word": false,
169
+ "special": true
170
+ },
171
+ "151350": {
172
+ "content": "<think>",
173
+ "lstrip": false,
174
+ "normalized": false,
175
+ "rstrip": false,
176
+ "single_word": false,
177
+ "special": false
178
+ },
179
+ "151351": {
180
+ "content": "</think>",
181
+ "lstrip": false,
182
+ "normalized": false,
183
+ "rstrip": false,
184
+ "single_word": false,
185
+ "special": false
186
+ },
187
+ "151352": {
188
+ "content": "<tool_call>",
189
+ "lstrip": false,
190
+ "normalized": false,
191
+ "rstrip": false,
192
+ "single_word": false,
193
+ "special": false
194
+ },
195
+ "151353": {
196
+ "content": "</tool_call>",
197
+ "lstrip": false,
198
+ "normalized": false,
199
+ "rstrip": false,
200
+ "single_word": false,
201
+ "special": false
202
+ },
203
+ "151354": {
204
+ "content": "<tool_response>",
205
+ "lstrip": false,
206
+ "normalized": false,
207
+ "rstrip": false,
208
+ "single_word": false,
209
+ "special": false
210
+ },
211
+ "151355": {
212
+ "content": "</tool_response>",
213
+ "lstrip": false,
214
+ "normalized": false,
215
+ "rstrip": false,
216
+ "single_word": false,
217
+ "special": false
218
+ },
219
+ "151356": {
220
+ "content": "<arg_key>",
221
+ "lstrip": false,
222
+ "normalized": false,
223
+ "rstrip": false,
224
+ "single_word": false,
225
+ "special": false
226
+ },
227
+ "151357": {
228
+ "content": "</arg_key>",
229
+ "lstrip": false,
230
+ "normalized": false,
231
+ "rstrip": false,
232
+ "single_word": false,
233
+ "special": false
234
+ },
235
+ "151358": {
236
+ "content": "<arg_value>",
237
+ "lstrip": false,
238
+ "normalized": false,
239
+ "rstrip": false,
240
+ "single_word": false,
241
+ "special": false
242
+ },
243
+ "151359": {
244
+ "content": "</arg_value>",
245
+ "lstrip": false,
246
+ "normalized": false,
247
+ "rstrip": false,
248
+ "single_word": false,
249
+ "special": false
250
+ },
251
+ "151360": {
252
+ "content": "/nothink",
253
+ "lstrip": false,
254
+ "normalized": false,
255
+ "rstrip": false,
256
+ "single_word": false,
257
+ "special": true
258
+ },
259
+ "151361": {
260
+ "content": "<|begin_of_box|>",
261
+ "lstrip": false,
262
+ "normalized": false,
263
+ "rstrip": false,
264
+ "single_word": false,
265
+ "special": false
266
+ },
267
+ "151362": {
268
+ "content": "<|end_of_box|>",
269
+ "lstrip": false,
270
+ "normalized": false,
271
+ "rstrip": false,
272
+ "single_word": false,
273
+ "special": false
274
+ },
275
+ "151363": {
276
+ "content": "<|image|>",
277
+ "lstrip": false,
278
+ "normalized": false,
279
+ "rstrip": false,
280
+ "single_word": false,
281
+ "special": false
282
+ },
283
+ "151364": {
284
+ "content": "<|video|>",
285
+ "lstrip": false,
286
+ "normalized": false,
287
+ "rstrip": false,
288
+ "single_word": false,
289
+ "special": false
290
+ }
291
+ },
292
+ "additional_special_tokens": [
293
+ "<|endoftext|>",
294
+ "[MASK]",
295
+ "[gMASK]",
296
+ "[sMASK]",
297
+ "<sop>",
298
+ "<eop>",
299
+ "<|system|>",
300
+ "<|user|>",
301
+ "<|assistant|>",
302
+ "<|observation|>",
303
+ "<|begin_of_image|>",
304
+ "<|end_of_image|>",
305
+ "<|begin_of_video|>",
306
+ "<|end_of_video|>",
307
+ "<|begin_of_audio|>",
308
+ "<|end_of_audio|>",
309
+ "<|begin_of_transcription|>",
310
+ "<|end_of_transcription|>",
311
+ "<|code_prefix|>",
312
+ "<|code_middle|>",
313
+ "<|code_suffix|>",
314
+ "/nothink"
315
+ ],
316
+ "clean_up_tokenization_spaces": false,
317
+ "do_lower_case": false,
318
+ "eos_token": "<|endoftext|>",
319
+ "extra_special_tokens": {},
320
+ "model_max_length": 128000,
321
+ "pad_token": "<|endoftext|>",
322
+ "padding_side": "left",
323
+ "remove_space": false,
324
+ "tokenizer_class": "PreTrainedTokenizerFast"
325
+ }