Xananthium commited on
Commit
9c42be4
·
verified ·
1 Parent(s): eaf3eb5

Publish qwen3.8-27b-hauhau-w8a8-vision-mtp with measured results and source credits

Browse files
Files changed (26) hide show
  1. .gitattributes +1 -0
  2. qwen3.8-27b-hauhau-w8a8-vision-mtp/README.md +17 -0
  3. qwen3.8-27b-hauhau-w8a8-vision-mtp/chat_template.jinja +173 -0
  4. qwen3.8-27b-hauhau-w8a8-vision-mtp/config.json +195 -0
  5. qwen3.8-27b-hauhau-w8a8-vision-mtp/conversion-receipt.json +410 -0
  6. qwen3.8-27b-hauhau-w8a8-vision-mtp/generation_config.json +13 -0
  7. qwen3.8-27b-hauhau-w8a8-vision-mtp/model-00000.safetensors +3 -0
  8. qwen3.8-27b-hauhau-w8a8-vision-mtp/model-00001.safetensors +3 -0
  9. qwen3.8-27b-hauhau-w8a8-vision-mtp/model-00002.safetensors +3 -0
  10. qwen3.8-27b-hauhau-w8a8-vision-mtp/model-00003.safetensors +3 -0
  11. qwen3.8-27b-hauhau-w8a8-vision-mtp/model-00004.safetensors +3 -0
  12. qwen3.8-27b-hauhau-w8a8-vision-mtp/model-00005.safetensors +3 -0
  13. qwen3.8-27b-hauhau-w8a8-vision-mtp/model-00006.safetensors +3 -0
  14. qwen3.8-27b-hauhau-w8a8-vision-mtp/model-00007.safetensors +3 -0
  15. qwen3.8-27b-hauhau-w8a8-vision-mtp/model-00008.safetensors +3 -0
  16. qwen3.8-27b-hauhau-w8a8-vision-mtp/model-00009.safetensors +3 -0
  17. qwen3.8-27b-hauhau-w8a8-vision-mtp/model-00010.safetensors +3 -0
  18. qwen3.8-27b-hauhau-w8a8-vision-mtp/model-00011.safetensors +3 -0
  19. qwen3.8-27b-hauhau-w8a8-vision-mtp/model-00012.safetensors +3 -0
  20. qwen3.8-27b-hauhau-w8a8-vision-mtp/model-00013.safetensors +3 -0
  21. qwen3.8-27b-hauhau-w8a8-vision-mtp/model.safetensors.index.json +0 -0
  22. qwen3.8-27b-hauhau-w8a8-vision-mtp/processor_config.json +60 -0
  23. qwen3.8-27b-hauhau-w8a8-vision-mtp/results/precision-comparison.json +382 -0
  24. qwen3.8-27b-hauhau-w8a8-vision-mtp/tokenizer.json +3 -0
  25. qwen3.8-27b-hauhau-w8a8-vision-mtp/tokenizer_config.json +33 -0
  26. qwen3.8-27b-hauhau-w8a8-vision-mtp/vision.safetensors +3 -0
.gitattributes CHANGED
@@ -46,3 +46,4 @@ qwen3.6-35b-hauhau-nvfp4-trt/tokenizer.json filter=lfs diff=lfs merge=lfs -text
46
  qwen3.6-35b-hauhau-reconstructed-bf16/tokenizer.json filter=lfs diff=lfs merge=lfs -text
47
  qwen3.6-35b-hauhau-w8a16-trt/tokenizer.json filter=lfs diff=lfs merge=lfs -text
48
  qwen3.6-35b-hauhau-w8a8-tq4-runtime/tokenizer.json filter=lfs diff=lfs merge=lfs -text
 
 
46
  qwen3.6-35b-hauhau-reconstructed-bf16/tokenizer.json filter=lfs diff=lfs merge=lfs -text
47
  qwen3.6-35b-hauhau-w8a16-trt/tokenizer.json filter=lfs diff=lfs merge=lfs -text
48
  qwen3.6-35b-hauhau-w8a8-tq4-runtime/tokenizer.json filter=lfs diff=lfs merge=lfs -text
49
+ qwen3.8-27b-hauhau-w8a8-vision-mtp/tokenizer.json filter=lfs diff=lfs merge=lfs -text
qwen3.8-27b-hauhau-w8a8-vision-mtp/README.md ADDED
@@ -0,0 +1,17 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ base_model: HauhauCS/Qwen3.8-27B-Uncensored-HauhauCS-Aggressive-MTP-GGUF
4
+ tags:
5
+ - local-inference
6
+ - quantization
7
+ - evaluation
8
+ ---
9
+
10
+ # qwen3.8-27b-hauhau-w8a8-vision-mtp
11
+
12
+ Complete local checkpoint published by Xananthium, 2026-10-06. Not evaluated in the 2026-10-06 matched precision comparison. The attached report describes other checkpoints, not scores for this artifact.
13
+
14
+ Source credit: [HauhauCS/Qwen3.8-27B-Uncensored-HauhauCS-Aggressive-MTP-GGUF](https://huggingface.co/HauhauCS/Qwen3.8-27B-Uncensored-HauhauCS-Aggressive-MTP-GGUF). Original base: [Qwen/Qwen3.8-27B](https://huggingface.co/Qwen/Qwen3.8-27B). Original model card is preserved in `README_ORIGINAL.md`; upstream licenses apply.
15
+
16
+ Measured results and limitations are in `results/precision-comparison.json`. These are small synthetic tests, not published cybersecurity benchmark scores. Untested artifacts have no measured scores.
17
+
qwen3.8-27b-hauhau-w8a8-vision-mtp/chat_template.jinja ADDED
@@ -0,0 +1,173 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- set image_count = namespace(value=0) %}
2
+ {%- set video_count = namespace(value=0) %}
3
+ {%- macro render_content(content, do_vision_count, is_system_content=false) %}
4
+ {%- if content is string %}
5
+ {{- content }}
6
+ {%- elif content is iterable and content is not mapping %}
7
+ {%- for item in content %}
8
+ {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
9
+ {%- if is_system_content %}
10
+ {{- raise_exception('System message cannot contain images.') }}
11
+ {%- endif %}
12
+ {%- if do_vision_count %}
13
+ {%- set image_count.value = image_count.value + 1 %}
14
+ {%- endif %}
15
+ {%- if add_vision_id %}
16
+ {{- 'Picture ' ~ image_count.value ~ ': ' }}
17
+ {%- endif %}
18
+ {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
19
+ {%- elif 'video' in item or item.type == 'video' %}
20
+ {%- if is_system_content %}
21
+ {{- raise_exception('System message cannot contain videos.') }}
22
+ {%- endif %}
23
+ {%- if do_vision_count %}
24
+ {%- set video_count.value = video_count.value + 1 %}
25
+ {%- endif %}
26
+ {%- if add_vision_id %}
27
+ {{- 'Video ' ~ video_count.value ~ ': ' }}
28
+ {%- endif %}
29
+ {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
30
+ {%- elif 'text' in item %}
31
+ {{- item.text }}
32
+ {%- else %}
33
+ {{- raise_exception('Unexpected item type in content.') }}
34
+ {%- endif %}
35
+ {%- endfor %}
36
+ {%- elif content is none or content is undefined %}
37
+ {{- '' }}
38
+ {%- else %}
39
+ {{- raise_exception('Unexpected content type.') }}
40
+ {%- endif %}
41
+ {%- endmacro %}
42
+ {%- if not messages %}
43
+ {{- raise_exception('No messages provided.') }}
44
+ {%- endif %}
45
+ {%- set reasoning_instructions = '' %}
46
+ {%- if enable_thinking is undefined or enable_thinking is true %}
47
+ {%- set resolved_reasoning_effort = reasoning_effort|default('xhigh') %}
48
+ {%- if resolved_reasoning_effort in ('high', 'max') %}
49
+ {%- set resolved_reasoning_effort = 'xhigh' %}
50
+ {%- endif %}
51
+ {%- if resolved_reasoning_effort not in ('xhigh', 'medium', 'low') %}
52
+ {{- raise_exception('Unexpected reasoning effort ' ~ reasoning_effort ~ '. Supported types are xhigh (default), medium, and low.') }}
53
+ {%- endif %}
54
+ {%- if resolved_reasoning_effort == 'xhigh' %}
55
+ {%- set reasoning_instructions = 'Reasoning effort is set to xhigh. Please think carefully through the task, validate key assumptions, consider plausible alternatives, and prioritize correctness, consistency, and clarity in the final answer.' %}
56
+ {%- elif resolved_reasoning_effort == 'low' %}
57
+ {%- set reasoning_instructions = 'Reasoning effort is set to low. Keep your thinking brief and focused, moving directly to the conclusion without unnecessary elaboration.' %}
58
+ {%- endif %}
59
+ {%- endif %}
60
+ {%- if tools and tools is iterable and tools is not mapping %}
61
+ {{- '<|im_start|>system\n' }}
62
+ {%- if reasoning_instructions %}
63
+ {{- reasoning_instructions + '\n\n' }}
64
+ {%- endif %}
65
+ {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
66
+ {%- for tool in tools %}
67
+ {{- "\n" }}
68
+ {{- tool | tojson }}
69
+ {%- endfor %}
70
+ {{- "\n</tools>" }}
71
+ {{- '\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>' }}
72
+ {%- if messages[0].role == 'system' %}
73
+ {%- set content = render_content(messages[0].content, false, true)|trim %}
74
+ {%- if content %}
75
+ {{- '\n\n' + content }}
76
+ {%- endif %}
77
+ {%- endif %}
78
+ {{- '<|im_end|>\n' }}
79
+ {%- else %}
80
+ {%- if messages[0].role == 'system' %}
81
+ {%- set content = render_content(messages[0].content, false, true)|trim %}
82
+ {%- if content %}
83
+ {{- '<|im_start|>system\n' + (reasoning_instructions + '\n\n' if reasoning_instructions else '') + content + '<|im_end|>\n' }}
84
+ {%- elif reasoning_instructions %}
85
+ {{- '<|im_start|>system\n' + reasoning_instructions + '<|im_end|>\n' }}
86
+ {%- endif %}
87
+ {%- elif reasoning_instructions %}
88
+ {{- '<|im_start|>system\n' + reasoning_instructions + '<|im_end|>\n' }}
89
+ {%- endif %}
90
+ {%- endif %}
91
+ {%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
92
+ {%- for message in messages[::-1] %}
93
+ {%- set index = (messages|length - 1) - loop.index0 %}
94
+ {%- if ns.multi_step_tool and message.role == "user" %}
95
+ {%- set content = render_content(message.content, false)|trim %}
96
+ {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
97
+ {%- set ns.multi_step_tool = false %}
98
+ {%- set ns.last_query_index = index %}
99
+ {%- endif %}
100
+ {%- endif %}
101
+ {%- endfor %}
102
+ {%- if ns.multi_step_tool %}
103
+ {{- raise_exception('No user query found in messages.') }}
104
+ {%- endif %}
105
+ {%- for message in messages %}
106
+ {%- set content = render_content(message.content, true)|trim %}
107
+ {%- if message.role == "system" %}
108
+ {%- if not loop.first %}
109
+ {{- raise_exception('System message must be at the beginning.') }}
110
+ {%- endif %}
111
+ {%- elif message.role == "user" %}
112
+ {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
113
+ {%- elif message.role == "assistant" %}
114
+ {%- set reasoning_content = '' %}
115
+ {%- if message.reasoning_content is string %}
116
+ {%- set reasoning_content = message.reasoning_content %}
117
+ {%- endif %}
118
+ {%- set reasoning_content = reasoning_content|trim %}
119
+ {%- if preserve_thinking is undefined or preserve_thinking is true or loop.index0 > ns.last_query_index %}
120
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
121
+ {%- else %}
122
+ {{- '<|im_start|>' + message.role + '\n' + content }}
123
+ {%- endif %}
124
+ {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
125
+ {%- for tool_call in message.tool_calls %}
126
+ {%- if tool_call.function is defined %}
127
+ {%- set tool_call = tool_call.function %}
128
+ {%- endif %}
129
+ {%- if loop.first %}
130
+ {%- if content|trim %}
131
+ {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
132
+ {%- else %}
133
+ {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
134
+ {%- endif %}
135
+ {%- else %}
136
+ {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
137
+ {%- endif %}
138
+ {%- if tool_call.arguments is defined and tool_call.arguments != '' %}
139
+ {%- for args_name, args_value in tool_call.arguments|items %}
140
+ {{- '<parameter=' + args_name + '>\n' }}
141
+ {%- set args_value = args_value | string if args_value is string else args_value | tojson | safe %}
142
+ {{- args_value }}
143
+ {{- '\n</parameter>\n' }}
144
+ {%- endfor %}
145
+ {%- endif %}
146
+ {{- '</function>\n</tool_call>' }}
147
+ {%- endfor %}
148
+ {%- endif %}
149
+ {{- '<|im_end|>\n' }}
150
+ {%- elif message.role == "tool" %}
151
+ {%- if loop.previtem and loop.previtem.role != "tool" %}
152
+ {{- '<|im_start|>user' }}
153
+ {%- endif %}
154
+ {{- '\n<tool_response>\n' }}
155
+ {{- content }}
156
+ {{- '\n</tool_response>' }}
157
+ {%- if not loop.last and loop.nextitem.role != "tool" %}
158
+ {{- '<|im_end|>\n' }}
159
+ {%- elif loop.last %}
160
+ {{- '<|im_end|>\n' }}
161
+ {%- endif %}
162
+ {%- else %}
163
+ {{- raise_exception('Unexpected message role.') }}
164
+ {%- endif %}
165
+ {%- endfor %}
166
+ {%- if add_generation_prompt %}
167
+ {{- '<|im_start|>assistant\n' }}
168
+ {%- if enable_thinking is defined and enable_thinking is false %}
169
+ {{- '<think>\n\n</think>\n\n' }}
170
+ {%- else %}
171
+ {{- '<think>\n' }}
172
+ {%- endif %}
173
+ {%- endif %}
qwen3.8-27b-hauhau-w8a8-vision-mtp/config.json ADDED
@@ -0,0 +1,195 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Qwen3_5ForConditionalGeneration"
4
+ ],
5
+ "dtype": "bfloat16",
6
+ "image_token_id": 248056,
7
+ "language_model_only": false,
8
+ "model_type": "qwen3_5",
9
+ "quantization_config": {
10
+ "config_groups": {
11
+ "config_group_0": {
12
+ "format": "int-quantized",
13
+ "input_activations": {
14
+ "actorder": null,
15
+ "block_structure": null,
16
+ "dynamic": true,
17
+ "group_size": null,
18
+ "num_bits": 8,
19
+ "observer": null,
20
+ "observer_kwargs": {},
21
+ "scale_dtype": null,
22
+ "strategy": "token",
23
+ "symmetric": true,
24
+ "type": "int",
25
+ "zp_dtype": null
26
+ },
27
+ "output_activations": null,
28
+ "targets": [
29
+ "Linear"
30
+ ],
31
+ "weights": {
32
+ "actorder": null,
33
+ "block_structure": null,
34
+ "dynamic": false,
35
+ "group_size": null,
36
+ "num_bits": 8,
37
+ "observer": "memoryless_minmax",
38
+ "observer_kwargs": {},
39
+ "scale_dtype": null,
40
+ "strategy": "channel",
41
+ "symmetric": true,
42
+ "type": "int",
43
+ "zp_dtype": null
44
+ }
45
+ }
46
+ },
47
+ "format": "int-quantized",
48
+ "global_compression_ratio": null,
49
+ "ignore": [
50
+ "lm_head",
51
+ "model.language_model.embed_tokens",
52
+ "re:.*visual.*",
53
+ "re:.*mtp.*",
54
+ "re:.*linear_attn.in_proj_[ab]$"
55
+ ],
56
+ "kv_cache_scheme": null,
57
+ "quant_method": "compressed-tensors",
58
+ "quantization_status": "compressed",
59
+ "transform_config": {},
60
+ "version": "0.17.1"
61
+ },
62
+ "text_config": {
63
+ "attention_bias": false,
64
+ "attention_dropout": 0.0,
65
+ "attn_output_gate": true,
66
+ "bos_token_id": 248044,
67
+ "dtype": "bfloat16",
68
+ "eos_token_id": 248044,
69
+ "full_attention_interval": 4,
70
+ "head_dim": 256,
71
+ "hidden_act": "silu",
72
+ "hidden_size": 5120,
73
+ "initializer_range": 0.02,
74
+ "intermediate_size": 17408,
75
+ "layer_types": [
76
+ "linear_attention",
77
+ "linear_attention",
78
+ "linear_attention",
79
+ "full_attention",
80
+ "linear_attention",
81
+ "linear_attention",
82
+ "linear_attention",
83
+ "full_attention",
84
+ "linear_attention",
85
+ "linear_attention",
86
+ "linear_attention",
87
+ "full_attention",
88
+ "linear_attention",
89
+ "linear_attention",
90
+ "linear_attention",
91
+ "full_attention",
92
+ "linear_attention",
93
+ "linear_attention",
94
+ "linear_attention",
95
+ "full_attention",
96
+ "linear_attention",
97
+ "linear_attention",
98
+ "linear_attention",
99
+ "full_attention",
100
+ "linear_attention",
101
+ "linear_attention",
102
+ "linear_attention",
103
+ "full_attention",
104
+ "linear_attention",
105
+ "linear_attention",
106
+ "linear_attention",
107
+ "full_attention",
108
+ "linear_attention",
109
+ "linear_attention",
110
+ "linear_attention",
111
+ "full_attention",
112
+ "linear_attention",
113
+ "linear_attention",
114
+ "linear_attention",
115
+ "full_attention",
116
+ "linear_attention",
117
+ "linear_attention",
118
+ "linear_attention",
119
+ "full_attention",
120
+ "linear_attention",
121
+ "linear_attention",
122
+ "linear_attention",
123
+ "full_attention",
124
+ "linear_attention",
125
+ "linear_attention",
126
+ "linear_attention",
127
+ "full_attention",
128
+ "linear_attention",
129
+ "linear_attention",
130
+ "linear_attention",
131
+ "full_attention",
132
+ "linear_attention",
133
+ "linear_attention",
134
+ "linear_attention",
135
+ "full_attention",
136
+ "linear_attention",
137
+ "linear_attention",
138
+ "linear_attention",
139
+ "full_attention"
140
+ ],
141
+ "linear_conv_kernel_dim": 4,
142
+ "linear_key_head_dim": 128,
143
+ "linear_num_key_heads": 16,
144
+ "linear_num_value_heads": 48,
145
+ "linear_value_head_dim": 128,
146
+ "mamba_ssm_dtype": "float32",
147
+ "max_position_embeddings": 262144,
148
+ "model_type": "qwen3_5_text",
149
+ "mtp_num_hidden_layers": 1,
150
+ "mtp_use_dedicated_embeddings": false,
151
+ "num_attention_heads": 24,
152
+ "num_hidden_layers": 64,
153
+ "num_key_value_heads": 4,
154
+ "output_gate_type": "swish",
155
+ "pad_token_id": null,
156
+ "partial_rotary_factor": 0.25,
157
+ "rms_norm_eps": 1e-06,
158
+ "rope_parameters": {
159
+ "mrope_interleaved": true,
160
+ "mrope_section": [
161
+ 11,
162
+ 11,
163
+ 10
164
+ ],
165
+ "partial_rotary_factor": 0.25,
166
+ "rope_theta": 10000000,
167
+ "rope_type": "default"
168
+ },
169
+ "tie_word_embeddings": false,
170
+ "use_cache": true,
171
+ "vocab_size": 248320
172
+ },
173
+ "tie_word_embeddings": false,
174
+ "transformers_version": "5.15.0",
175
+ "video_token_id": 248057,
176
+ "vision_config": {
177
+ "deepstack_visual_indexes": [],
178
+ "depth": 27,
179
+ "dtype": "bfloat16",
180
+ "hidden_act": "gelu_pytorch_tanh",
181
+ "hidden_size": 1152,
182
+ "in_channels": 3,
183
+ "initializer_range": 0.02,
184
+ "intermediate_size": 4304,
185
+ "model_type": "qwen3_5_vision",
186
+ "num_heads": 16,
187
+ "num_position_embeddings": 2304,
188
+ "out_hidden_size": 5120,
189
+ "patch_size": 16,
190
+ "spatial_merge_size": 2,
191
+ "temporal_patch_size": 2
192
+ },
193
+ "vision_end_token_id": 248054,
194
+ "vision_start_token_id": 248053
195
+ }
qwen3.8-27b-hauhau-w8a8-vision-mtp/conversion-receipt.json ADDED
@@ -0,0 +1,410 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "source_repo": "HauhauCS/Qwen3.8-27B-Uncensored-HauhauCS-Aggressive-MTP-GGUF",
3
+ "source_file": "Qwen3.8-27B-Uncensored-HauhauCS-Aggressive-Q8_K_P.gguf",
4
+ "precision": "INT8 weights, dynamic per-token INT8 activations; embeddings, lm_head, vision, MTP, small a/b projections BF16",
5
+ "quantized_linears": [
6
+ "model.language_model.layers.0.linear_attn.out_proj.weight",
7
+ "model.language_model.layers.0.linear_attn.in_proj_qkv.weight",
8
+ "model.language_model.layers.0.linear_attn.in_proj_z.weight",
9
+ "model.language_model.layers.0.mlp.gate_proj.weight",
10
+ "model.language_model.layers.0.mlp.up_proj.weight",
11
+ "model.language_model.layers.0.mlp.down_proj.weight",
12
+ "model.language_model.layers.1.linear_attn.out_proj.weight",
13
+ "model.language_model.layers.1.linear_attn.in_proj_qkv.weight",
14
+ "model.language_model.layers.1.linear_attn.in_proj_z.weight",
15
+ "model.language_model.layers.1.mlp.gate_proj.weight",
16
+ "model.language_model.layers.1.mlp.up_proj.weight",
17
+ "model.language_model.layers.1.mlp.down_proj.weight",
18
+ "model.language_model.layers.2.linear_attn.out_proj.weight",
19
+ "model.language_model.layers.2.linear_attn.in_proj_qkv.weight",
20
+ "model.language_model.layers.2.linear_attn.in_proj_z.weight",
21
+ "model.language_model.layers.2.mlp.gate_proj.weight",
22
+ "model.language_model.layers.2.mlp.up_proj.weight",
23
+ "model.language_model.layers.2.mlp.down_proj.weight",
24
+ "model.language_model.layers.3.self_attn.q_proj.weight",
25
+ "model.language_model.layers.3.self_attn.k_proj.weight",
26
+ "model.language_model.layers.3.self_attn.v_proj.weight",
27
+ "model.language_model.layers.3.self_attn.o_proj.weight",
28
+ "model.language_model.layers.3.mlp.gate_proj.weight",
29
+ "model.language_model.layers.3.mlp.up_proj.weight",
30
+ "model.language_model.layers.3.mlp.down_proj.weight",
31
+ "model.language_model.layers.4.linear_attn.out_proj.weight",
32
+ "model.language_model.layers.4.linear_attn.in_proj_qkv.weight",
33
+ "model.language_model.layers.4.linear_attn.in_proj_z.weight",
34
+ "model.language_model.layers.4.mlp.gate_proj.weight",
35
+ "model.language_model.layers.4.mlp.up_proj.weight",
36
+ "model.language_model.layers.4.mlp.down_proj.weight",
37
+ "model.language_model.layers.5.linear_attn.out_proj.weight",
38
+ "model.language_model.layers.5.linear_attn.in_proj_qkv.weight",
39
+ "model.language_model.layers.5.linear_attn.in_proj_z.weight",
40
+ "model.language_model.layers.5.mlp.gate_proj.weight",
41
+ "model.language_model.layers.5.mlp.up_proj.weight",
42
+ "model.language_model.layers.5.mlp.down_proj.weight",
43
+ "model.language_model.layers.6.linear_attn.out_proj.weight",
44
+ "model.language_model.layers.6.linear_attn.in_proj_qkv.weight",
45
+ "model.language_model.layers.6.linear_attn.in_proj_z.weight",
46
+ "model.language_model.layers.6.mlp.gate_proj.weight",
47
+ "model.language_model.layers.6.mlp.up_proj.weight",
48
+ "model.language_model.layers.6.mlp.down_proj.weight",
49
+ "model.language_model.layers.7.self_attn.q_proj.weight",
50
+ "model.language_model.layers.7.self_attn.k_proj.weight",
51
+ "model.language_model.layers.7.self_attn.v_proj.weight",
52
+ "model.language_model.layers.7.self_attn.o_proj.weight",
53
+ "model.language_model.layers.7.mlp.gate_proj.weight",
54
+ "model.language_model.layers.7.mlp.up_proj.weight",
55
+ "model.language_model.layers.7.mlp.down_proj.weight",
56
+ "model.language_model.layers.8.linear_attn.out_proj.weight",
57
+ "model.language_model.layers.8.linear_attn.in_proj_qkv.weight",
58
+ "model.language_model.layers.8.linear_attn.in_proj_z.weight",
59
+ "model.language_model.layers.8.mlp.gate_proj.weight",
60
+ "model.language_model.layers.8.mlp.up_proj.weight",
61
+ "model.language_model.layers.8.mlp.down_proj.weight",
62
+ "model.language_model.layers.9.linear_attn.out_proj.weight",
63
+ "model.language_model.layers.9.linear_attn.in_proj_qkv.weight",
64
+ "model.language_model.layers.9.linear_attn.in_proj_z.weight",
65
+ "model.language_model.layers.9.mlp.gate_proj.weight",
66
+ "model.language_model.layers.9.mlp.up_proj.weight",
67
+ "model.language_model.layers.9.mlp.down_proj.weight",
68
+ "model.language_model.layers.10.linear_attn.out_proj.weight",
69
+ "model.language_model.layers.10.linear_attn.in_proj_qkv.weight",
70
+ "model.language_model.layers.10.linear_attn.in_proj_z.weight",
71
+ "model.language_model.layers.10.mlp.gate_proj.weight",
72
+ "model.language_model.layers.10.mlp.up_proj.weight",
73
+ "model.language_model.layers.10.mlp.down_proj.weight",
74
+ "model.language_model.layers.11.self_attn.q_proj.weight",
75
+ "model.language_model.layers.11.self_attn.k_proj.weight",
76
+ "model.language_model.layers.11.self_attn.v_proj.weight",
77
+ "model.language_model.layers.11.self_attn.o_proj.weight",
78
+ "model.language_model.layers.11.mlp.gate_proj.weight",
79
+ "model.language_model.layers.11.mlp.up_proj.weight",
80
+ "model.language_model.layers.11.mlp.down_proj.weight",
81
+ "model.language_model.layers.12.linear_attn.out_proj.weight",
82
+ "model.language_model.layers.12.linear_attn.in_proj_qkv.weight",
83
+ "model.language_model.layers.12.linear_attn.in_proj_z.weight",
84
+ "model.language_model.layers.12.mlp.gate_proj.weight",
85
+ "model.language_model.layers.12.mlp.up_proj.weight",
86
+ "model.language_model.layers.12.mlp.down_proj.weight",
87
+ "model.language_model.layers.13.linear_attn.out_proj.weight",
88
+ "model.language_model.layers.13.linear_attn.in_proj_qkv.weight",
89
+ "model.language_model.layers.13.linear_attn.in_proj_z.weight",
90
+ "model.language_model.layers.13.mlp.gate_proj.weight",
91
+ "model.language_model.layers.13.mlp.up_proj.weight",
92
+ "model.language_model.layers.13.mlp.down_proj.weight",
93
+ "model.language_model.layers.14.linear_attn.out_proj.weight",
94
+ "model.language_model.layers.14.linear_attn.in_proj_qkv.weight",
95
+ "model.language_model.layers.14.linear_attn.in_proj_z.weight",
96
+ "model.language_model.layers.14.mlp.gate_proj.weight",
97
+ "model.language_model.layers.14.mlp.up_proj.weight",
98
+ "model.language_model.layers.14.mlp.down_proj.weight",
99
+ "model.language_model.layers.15.self_attn.q_proj.weight",
100
+ "model.language_model.layers.15.self_attn.k_proj.weight",
101
+ "model.language_model.layers.15.self_attn.v_proj.weight",
102
+ "model.language_model.layers.15.self_attn.o_proj.weight",
103
+ "model.language_model.layers.15.mlp.gate_proj.weight",
104
+ "model.language_model.layers.15.mlp.up_proj.weight",
105
+ "model.language_model.layers.15.mlp.down_proj.weight",
106
+ "model.language_model.layers.16.linear_attn.out_proj.weight",
107
+ "model.language_model.layers.16.linear_attn.in_proj_qkv.weight",
108
+ "model.language_model.layers.16.linear_attn.in_proj_z.weight",
109
+ "model.language_model.layers.16.mlp.gate_proj.weight",
110
+ "model.language_model.layers.16.mlp.up_proj.weight",
111
+ "model.language_model.layers.16.mlp.down_proj.weight",
112
+ "model.language_model.layers.17.linear_attn.out_proj.weight",
113
+ "model.language_model.layers.17.linear_attn.in_proj_qkv.weight",
114
+ "model.language_model.layers.17.linear_attn.in_proj_z.weight",
115
+ "model.language_model.layers.17.mlp.gate_proj.weight",
116
+ "model.language_model.layers.17.mlp.up_proj.weight",
117
+ "model.language_model.layers.17.mlp.down_proj.weight",
118
+ "model.language_model.layers.18.linear_attn.out_proj.weight",
119
+ "model.language_model.layers.18.linear_attn.in_proj_qkv.weight",
120
+ "model.language_model.layers.18.linear_attn.in_proj_z.weight",
121
+ "model.language_model.layers.18.mlp.gate_proj.weight",
122
+ "model.language_model.layers.18.mlp.up_proj.weight",
123
+ "model.language_model.layers.18.mlp.down_proj.weight",
124
+ "model.language_model.layers.19.self_attn.q_proj.weight",
125
+ "model.language_model.layers.19.self_attn.k_proj.weight",
126
+ "model.language_model.layers.19.self_attn.v_proj.weight",
127
+ "model.language_model.layers.19.self_attn.o_proj.weight",
128
+ "model.language_model.layers.19.mlp.gate_proj.weight",
129
+ "model.language_model.layers.19.mlp.up_proj.weight",
130
+ "model.language_model.layers.19.mlp.down_proj.weight",
131
+ "model.language_model.layers.20.linear_attn.out_proj.weight",
132
+ "model.language_model.layers.20.linear_attn.in_proj_qkv.weight",
133
+ "model.language_model.layers.20.linear_attn.in_proj_z.weight",
134
+ "model.language_model.layers.20.mlp.gate_proj.weight",
135
+ "model.language_model.layers.20.mlp.up_proj.weight",
136
+ "model.language_model.layers.20.mlp.down_proj.weight",
137
+ "model.language_model.layers.21.linear_attn.out_proj.weight",
138
+ "model.language_model.layers.21.linear_attn.in_proj_qkv.weight",
139
+ "model.language_model.layers.21.linear_attn.in_proj_z.weight",
140
+ "model.language_model.layers.21.mlp.gate_proj.weight",
141
+ "model.language_model.layers.21.mlp.up_proj.weight",
142
+ "model.language_model.layers.21.mlp.down_proj.weight",
143
+ "model.language_model.layers.22.linear_attn.out_proj.weight",
144
+ "model.language_model.layers.22.linear_attn.in_proj_qkv.weight",
145
+ "model.language_model.layers.22.linear_attn.in_proj_z.weight",
146
+ "model.language_model.layers.22.mlp.gate_proj.weight",
147
+ "model.language_model.layers.22.mlp.up_proj.weight",
148
+ "model.language_model.layers.22.mlp.down_proj.weight",
149
+ "model.language_model.layers.23.self_attn.q_proj.weight",
150
+ "model.language_model.layers.23.self_attn.k_proj.weight",
151
+ "model.language_model.layers.23.self_attn.v_proj.weight",
152
+ "model.language_model.layers.23.self_attn.o_proj.weight",
153
+ "model.language_model.layers.23.mlp.gate_proj.weight",
154
+ "model.language_model.layers.23.mlp.up_proj.weight",
155
+ "model.language_model.layers.23.mlp.down_proj.weight",
156
+ "model.language_model.layers.24.linear_attn.out_proj.weight",
157
+ "model.language_model.layers.24.linear_attn.in_proj_qkv.weight",
158
+ "model.language_model.layers.24.linear_attn.in_proj_z.weight",
159
+ "model.language_model.layers.24.mlp.gate_proj.weight",
160
+ "model.language_model.layers.24.mlp.up_proj.weight",
161
+ "model.language_model.layers.24.mlp.down_proj.weight",
162
+ "model.language_model.layers.25.linear_attn.out_proj.weight",
163
+ "model.language_model.layers.25.linear_attn.in_proj_qkv.weight",
164
+ "model.language_model.layers.25.linear_attn.in_proj_z.weight",
165
+ "model.language_model.layers.25.mlp.gate_proj.weight",
166
+ "model.language_model.layers.25.mlp.up_proj.weight",
167
+ "model.language_model.layers.25.mlp.down_proj.weight",
168
+ "model.language_model.layers.26.linear_attn.out_proj.weight",
169
+ "model.language_model.layers.26.linear_attn.in_proj_qkv.weight",
170
+ "model.language_model.layers.26.linear_attn.in_proj_z.weight",
171
+ "model.language_model.layers.26.mlp.gate_proj.weight",
172
+ "model.language_model.layers.26.mlp.up_proj.weight",
173
+ "model.language_model.layers.26.mlp.down_proj.weight",
174
+ "model.language_model.layers.27.self_attn.q_proj.weight",
175
+ "model.language_model.layers.27.self_attn.k_proj.weight",
176
+ "model.language_model.layers.27.self_attn.v_proj.weight",
177
+ "model.language_model.layers.27.self_attn.o_proj.weight",
178
+ "model.language_model.layers.27.mlp.gate_proj.weight",
179
+ "model.language_model.layers.27.mlp.up_proj.weight",
180
+ "model.language_model.layers.27.mlp.down_proj.weight",
181
+ "model.language_model.layers.28.linear_attn.out_proj.weight",
182
+ "model.language_model.layers.28.linear_attn.in_proj_qkv.weight",
183
+ "model.language_model.layers.28.linear_attn.in_proj_z.weight",
184
+ "model.language_model.layers.28.mlp.gate_proj.weight",
185
+ "model.language_model.layers.28.mlp.up_proj.weight",
186
+ "model.language_model.layers.28.mlp.down_proj.weight",
187
+ "model.language_model.layers.29.linear_attn.out_proj.weight",
188
+ "model.language_model.layers.29.linear_attn.in_proj_qkv.weight",
189
+ "model.language_model.layers.29.linear_attn.in_proj_z.weight",
190
+ "model.language_model.layers.29.mlp.gate_proj.weight",
191
+ "model.language_model.layers.29.mlp.up_proj.weight",
192
+ "model.language_model.layers.29.mlp.down_proj.weight",
193
+ "model.language_model.layers.30.linear_attn.out_proj.weight",
194
+ "model.language_model.layers.30.linear_attn.in_proj_qkv.weight",
195
+ "model.language_model.layers.30.linear_attn.in_proj_z.weight",
196
+ "model.language_model.layers.30.mlp.gate_proj.weight",
197
+ "model.language_model.layers.30.mlp.up_proj.weight",
198
+ "model.language_model.layers.30.mlp.down_proj.weight",
199
+ "model.language_model.layers.31.self_attn.q_proj.weight",
200
+ "model.language_model.layers.31.self_attn.k_proj.weight",
201
+ "model.language_model.layers.31.self_attn.v_proj.weight",
202
+ "model.language_model.layers.31.self_attn.o_proj.weight",
203
+ "model.language_model.layers.31.mlp.gate_proj.weight",
204
+ "model.language_model.layers.31.mlp.up_proj.weight",
205
+ "model.language_model.layers.31.mlp.down_proj.weight",
206
+ "model.language_model.layers.32.linear_attn.out_proj.weight",
207
+ "model.language_model.layers.32.linear_attn.in_proj_qkv.weight",
208
+ "model.language_model.layers.32.linear_attn.in_proj_z.weight",
209
+ "model.language_model.layers.32.mlp.gate_proj.weight",
210
+ "model.language_model.layers.32.mlp.up_proj.weight",
211
+ "model.language_model.layers.32.mlp.down_proj.weight",
212
+ "model.language_model.layers.33.linear_attn.out_proj.weight",
213
+ "model.language_model.layers.33.linear_attn.in_proj_qkv.weight",
214
+ "model.language_model.layers.33.linear_attn.in_proj_z.weight",
215
+ "model.language_model.layers.33.mlp.gate_proj.weight",
216
+ "model.language_model.layers.33.mlp.up_proj.weight",
217
+ "model.language_model.layers.33.mlp.down_proj.weight",
218
+ "model.language_model.layers.34.linear_attn.out_proj.weight",
219
+ "model.language_model.layers.34.linear_attn.in_proj_qkv.weight",
220
+ "model.language_model.layers.34.linear_attn.in_proj_z.weight",
221
+ "model.language_model.layers.34.mlp.gate_proj.weight",
222
+ "model.language_model.layers.34.mlp.up_proj.weight",
223
+ "model.language_model.layers.34.mlp.down_proj.weight",
224
+ "model.language_model.layers.35.self_attn.q_proj.weight",
225
+ "model.language_model.layers.35.self_attn.k_proj.weight",
226
+ "model.language_model.layers.35.self_attn.v_proj.weight",
227
+ "model.language_model.layers.35.self_attn.o_proj.weight",
228
+ "model.language_model.layers.35.mlp.gate_proj.weight",
229
+ "model.language_model.layers.35.mlp.up_proj.weight",
230
+ "model.language_model.layers.35.mlp.down_proj.weight",
231
+ "model.language_model.layers.36.linear_attn.out_proj.weight",
232
+ "model.language_model.layers.36.linear_attn.in_proj_qkv.weight",
233
+ "model.language_model.layers.36.linear_attn.in_proj_z.weight",
234
+ "model.language_model.layers.36.mlp.gate_proj.weight",
235
+ "model.language_model.layers.36.mlp.up_proj.weight",
236
+ "model.language_model.layers.36.mlp.down_proj.weight",
237
+ "model.language_model.layers.37.linear_attn.out_proj.weight",
238
+ "model.language_model.layers.37.linear_attn.in_proj_qkv.weight",
239
+ "model.language_model.layers.37.linear_attn.in_proj_z.weight",
240
+ "model.language_model.layers.37.mlp.gate_proj.weight",
241
+ "model.language_model.layers.37.mlp.up_proj.weight",
242
+ "model.language_model.layers.37.mlp.down_proj.weight",
243
+ "model.language_model.layers.38.linear_attn.out_proj.weight",
244
+ "model.language_model.layers.38.linear_attn.in_proj_qkv.weight",
245
+ "model.language_model.layers.38.linear_attn.in_proj_z.weight",
246
+ "model.language_model.layers.38.mlp.gate_proj.weight",
247
+ "model.language_model.layers.38.mlp.up_proj.weight",
248
+ "model.language_model.layers.38.mlp.down_proj.weight",
249
+ "model.language_model.layers.39.self_attn.q_proj.weight",
250
+ "model.language_model.layers.39.self_attn.k_proj.weight",
251
+ "model.language_model.layers.39.self_attn.v_proj.weight",
252
+ "model.language_model.layers.39.self_attn.o_proj.weight",
253
+ "model.language_model.layers.39.mlp.gate_proj.weight",
254
+ "model.language_model.layers.39.mlp.up_proj.weight",
255
+ "model.language_model.layers.39.mlp.down_proj.weight",
256
+ "model.language_model.layers.40.linear_attn.out_proj.weight",
257
+ "model.language_model.layers.40.linear_attn.in_proj_qkv.weight",
258
+ "model.language_model.layers.40.linear_attn.in_proj_z.weight",
259
+ "model.language_model.layers.40.mlp.gate_proj.weight",
260
+ "model.language_model.layers.40.mlp.up_proj.weight",
261
+ "model.language_model.layers.40.mlp.down_proj.weight",
262
+ "model.language_model.layers.41.linear_attn.out_proj.weight",
263
+ "model.language_model.layers.41.linear_attn.in_proj_qkv.weight",
264
+ "model.language_model.layers.41.linear_attn.in_proj_z.weight",
265
+ "model.language_model.layers.41.mlp.gate_proj.weight",
266
+ "model.language_model.layers.41.mlp.up_proj.weight",
267
+ "model.language_model.layers.41.mlp.down_proj.weight",
268
+ "model.language_model.layers.42.linear_attn.out_proj.weight",
269
+ "model.language_model.layers.42.linear_attn.in_proj_qkv.weight",
270
+ "model.language_model.layers.42.linear_attn.in_proj_z.weight",
271
+ "model.language_model.layers.42.mlp.gate_proj.weight",
272
+ "model.language_model.layers.42.mlp.up_proj.weight",
273
+ "model.language_model.layers.42.mlp.down_proj.weight",
274
+ "model.language_model.layers.43.self_attn.q_proj.weight",
275
+ "model.language_model.layers.43.self_attn.k_proj.weight",
276
+ "model.language_model.layers.43.self_attn.v_proj.weight",
277
+ "model.language_model.layers.43.self_attn.o_proj.weight",
278
+ "model.language_model.layers.43.mlp.gate_proj.weight",
279
+ "model.language_model.layers.43.mlp.up_proj.weight",
280
+ "model.language_model.layers.43.mlp.down_proj.weight",
281
+ "model.language_model.layers.44.linear_attn.out_proj.weight",
282
+ "model.language_model.layers.44.linear_attn.in_proj_qkv.weight",
283
+ "model.language_model.layers.44.linear_attn.in_proj_z.weight",
284
+ "model.language_model.layers.44.mlp.gate_proj.weight",
285
+ "model.language_model.layers.44.mlp.up_proj.weight",
286
+ "model.language_model.layers.44.mlp.down_proj.weight",
287
+ "model.language_model.layers.45.linear_attn.out_proj.weight",
288
+ "model.language_model.layers.45.linear_attn.in_proj_qkv.weight",
289
+ "model.language_model.layers.45.linear_attn.in_proj_z.weight",
290
+ "model.language_model.layers.45.mlp.gate_proj.weight",
291
+ "model.language_model.layers.45.mlp.up_proj.weight",
292
+ "model.language_model.layers.45.mlp.down_proj.weight",
293
+ "model.language_model.layers.46.linear_attn.out_proj.weight",
294
+ "model.language_model.layers.46.linear_attn.in_proj_qkv.weight",
295
+ "model.language_model.layers.46.linear_attn.in_proj_z.weight",
296
+ "model.language_model.layers.46.mlp.gate_proj.weight",
297
+ "model.language_model.layers.46.mlp.up_proj.weight",
298
+ "model.language_model.layers.46.mlp.down_proj.weight",
299
+ "model.language_model.layers.47.self_attn.q_proj.weight",
300
+ "model.language_model.layers.47.self_attn.k_proj.weight",
301
+ "model.language_model.layers.47.self_attn.v_proj.weight",
302
+ "model.language_model.layers.47.self_attn.o_proj.weight",
303
+ "model.language_model.layers.47.mlp.gate_proj.weight",
304
+ "model.language_model.layers.47.mlp.up_proj.weight",
305
+ "model.language_model.layers.47.mlp.down_proj.weight",
306
+ "model.language_model.layers.48.linear_attn.out_proj.weight",
307
+ "model.language_model.layers.48.linear_attn.in_proj_qkv.weight",
308
+ "model.language_model.layers.48.linear_attn.in_proj_z.weight",
309
+ "model.language_model.layers.48.mlp.gate_proj.weight",
310
+ "model.language_model.layers.48.mlp.up_proj.weight",
311
+ "model.language_model.layers.48.mlp.down_proj.weight",
312
+ "model.language_model.layers.49.linear_attn.out_proj.weight",
313
+ "model.language_model.layers.49.linear_attn.in_proj_qkv.weight",
314
+ "model.language_model.layers.49.linear_attn.in_proj_z.weight",
315
+ "model.language_model.layers.49.mlp.gate_proj.weight",
316
+ "model.language_model.layers.49.mlp.up_proj.weight",
317
+ "model.language_model.layers.49.mlp.down_proj.weight",
318
+ "model.language_model.layers.50.linear_attn.out_proj.weight",
319
+ "model.language_model.layers.50.linear_attn.in_proj_qkv.weight",
320
+ "model.language_model.layers.50.linear_attn.in_proj_z.weight",
321
+ "model.language_model.layers.50.mlp.gate_proj.weight",
322
+ "model.language_model.layers.50.mlp.up_proj.weight",
323
+ "model.language_model.layers.50.mlp.down_proj.weight",
324
+ "model.language_model.layers.51.self_attn.q_proj.weight",
325
+ "model.language_model.layers.51.self_attn.k_proj.weight",
326
+ "model.language_model.layers.51.self_attn.v_proj.weight",
327
+ "model.language_model.layers.51.self_attn.o_proj.weight",
328
+ "model.language_model.layers.51.mlp.gate_proj.weight",
329
+ "model.language_model.layers.51.mlp.up_proj.weight",
330
+ "model.language_model.layers.51.mlp.down_proj.weight",
331
+ "model.language_model.layers.52.linear_attn.out_proj.weight",
332
+ "model.language_model.layers.52.linear_attn.in_proj_qkv.weight",
333
+ "model.language_model.layers.52.linear_attn.in_proj_z.weight",
334
+ "model.language_model.layers.52.mlp.gate_proj.weight",
335
+ "model.language_model.layers.52.mlp.up_proj.weight",
336
+ "model.language_model.layers.52.mlp.down_proj.weight",
337
+ "model.language_model.layers.53.linear_attn.out_proj.weight",
338
+ "model.language_model.layers.53.linear_attn.in_proj_qkv.weight",
339
+ "model.language_model.layers.53.linear_attn.in_proj_z.weight",
340
+ "model.language_model.layers.53.mlp.gate_proj.weight",
341
+ "model.language_model.layers.53.mlp.up_proj.weight",
342
+ "model.language_model.layers.53.mlp.down_proj.weight",
343
+ "model.language_model.layers.54.linear_attn.out_proj.weight",
344
+ "model.language_model.layers.54.linear_attn.in_proj_qkv.weight",
345
+ "model.language_model.layers.54.linear_attn.in_proj_z.weight",
346
+ "model.language_model.layers.54.mlp.gate_proj.weight",
347
+ "model.language_model.layers.54.mlp.up_proj.weight",
348
+ "model.language_model.layers.54.mlp.down_proj.weight",
349
+ "model.language_model.layers.55.self_attn.q_proj.weight",
350
+ "model.language_model.layers.55.self_attn.k_proj.weight",
351
+ "model.language_model.layers.55.self_attn.v_proj.weight",
352
+ "model.language_model.layers.55.self_attn.o_proj.weight",
353
+ "model.language_model.layers.55.mlp.gate_proj.weight",
354
+ "model.language_model.layers.55.mlp.up_proj.weight",
355
+ "model.language_model.layers.55.mlp.down_proj.weight",
356
+ "model.language_model.layers.56.linear_attn.out_proj.weight",
357
+ "model.language_model.layers.56.linear_attn.in_proj_qkv.weight",
358
+ "model.language_model.layers.56.linear_attn.in_proj_z.weight",
359
+ "model.language_model.layers.56.mlp.gate_proj.weight",
360
+ "model.language_model.layers.56.mlp.up_proj.weight",
361
+ "model.language_model.layers.56.mlp.down_proj.weight",
362
+ "model.language_model.layers.57.linear_attn.out_proj.weight",
363
+ "model.language_model.layers.57.linear_attn.in_proj_qkv.weight",
364
+ "model.language_model.layers.57.linear_attn.in_proj_z.weight",
365
+ "model.language_model.layers.57.mlp.gate_proj.weight",
366
+ "model.language_model.layers.57.mlp.up_proj.weight",
367
+ "model.language_model.layers.57.mlp.down_proj.weight",
368
+ "model.language_model.layers.58.linear_attn.out_proj.weight",
369
+ "model.language_model.layers.58.linear_attn.in_proj_qkv.weight",
370
+ "model.language_model.layers.58.linear_attn.in_proj_z.weight",
371
+ "model.language_model.layers.58.mlp.gate_proj.weight",
372
+ "model.language_model.layers.58.mlp.up_proj.weight",
373
+ "model.language_model.layers.58.mlp.down_proj.weight",
374
+ "model.language_model.layers.59.self_attn.q_proj.weight",
375
+ "model.language_model.layers.59.self_attn.k_proj.weight",
376
+ "model.language_model.layers.59.self_attn.v_proj.weight",
377
+ "model.language_model.layers.59.self_attn.o_proj.weight",
378
+ "model.language_model.layers.59.mlp.gate_proj.weight",
379
+ "model.language_model.layers.59.mlp.up_proj.weight",
380
+ "model.language_model.layers.59.mlp.down_proj.weight",
381
+ "model.language_model.layers.60.linear_attn.out_proj.weight",
382
+ "model.language_model.layers.60.linear_attn.in_proj_qkv.weight",
383
+ "model.language_model.layers.60.linear_attn.in_proj_z.weight",
384
+ "model.language_model.layers.60.mlp.gate_proj.weight",
385
+ "model.language_model.layers.60.mlp.up_proj.weight",
386
+ "model.language_model.layers.60.mlp.down_proj.weight",
387
+ "model.language_model.layers.61.linear_attn.out_proj.weight",
388
+ "model.language_model.layers.61.linear_attn.in_proj_qkv.weight",
389
+ "model.language_model.layers.61.linear_attn.in_proj_z.weight",
390
+ "model.language_model.layers.61.mlp.gate_proj.weight",
391
+ "model.language_model.layers.61.mlp.up_proj.weight",
392
+ "model.language_model.layers.61.mlp.down_proj.weight",
393
+ "model.language_model.layers.62.linear_attn.out_proj.weight",
394
+ "model.language_model.layers.62.linear_attn.in_proj_qkv.weight",
395
+ "model.language_model.layers.62.linear_attn.in_proj_z.weight",
396
+ "model.language_model.layers.62.mlp.gate_proj.weight",
397
+ "model.language_model.layers.62.mlp.up_proj.weight",
398
+ "model.language_model.layers.62.mlp.down_proj.weight",
399
+ "model.language_model.layers.63.self_attn.q_proj.weight",
400
+ "model.language_model.layers.63.self_attn.k_proj.weight",
401
+ "model.language_model.layers.63.self_attn.v_proj.weight",
402
+ "model.language_model.layers.63.self_attn.o_proj.weight",
403
+ "model.language_model.layers.63.mlp.gate_proj.weight",
404
+ "model.language_model.layers.63.mlp.up_proj.weight",
405
+ "model.language_model.layers.63.mlp.down_proj.weight"
406
+ ],
407
+ "mtp": "matching Hauhau head retained in BF16",
408
+ "elapsed_seconds": 124.82400365800004,
409
+ "complete": true
410
+ }
qwen3.8-27b-hauhau-w8a8-vision-mtp/generation_config.json ADDED
@@ -0,0 +1,13 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token_id": 248044,
3
+ "do_sample": true,
4
+ "eos_token_id": [
5
+ 248046,
6
+ 248044
7
+ ],
8
+ "pad_token_id": 248044,
9
+ "temperature": 1.0,
10
+ "top_k": 20,
11
+ "top_p": 0.95,
12
+ "transformers_version": "5.15.0"
13
+ }
qwen3.8-27b-hauhau-w8a8-vision-mtp/model-00000.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9371b3c01e6b8fc586c9da33d0ae1b29d206cb930e45847f02c3626caf121ecb
3
+ size 2542796928
qwen3.8-27b-hauhau-w8a8-vision-mtp/model-00001.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:65df23e620a8371d8a3b791947d1a119dc3355f006a115f3d3deb1ff46972159
3
+ size 2023623272
qwen3.8-27b-hauhau-w8a8-vision-mtp/model-00002.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9cb14086fd0460295b1f7e4d48e98232eb5fbc3268a0662fd6562564367b7963
3
+ size 2087464624
qwen3.8-27b-hauhau-w8a8-vision-mtp/model-00003.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d3186cfca81f36c644da237725556d0a2bd3d9b535baef2080a5658cce290cfa
3
+ size 2017309632
qwen3.8-27b-hauhau-w8a8-vision-mtp/model-00004.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1418e7e4176e3ef22fc8673123c20ac8dc84686ecc863bfa8468f49064cca453
3
+ size 2082220088
qwen3.8-27b-hauhau-w8a8-vision-mtp/model-00005.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:935660b5cce2276b74fb40dda107f1bcd8fc0abe87dccb60f22d0e43814c6ab4
3
+ size 2086456824
qwen3.8-27b-hauhau-w8a8-vision-mtp/model-00006.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d51d3c24a2327815aa58214a7a45d95821717f237f556477fe74ac50d224f5e8
3
+ size 2012089576
qwen3.8-27b-hauhau-w8a8-vision-mtp/model-00007.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d51ce7e051480e36218f8dcce7db3199643b85c79c6e977d1fbc90230fc56b69
3
+ size 2087464728
qwen3.8-27b-hauhau-w8a8-vision-mtp/model-00008.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b6d56d06507d1d0e9c77e37c3c7036cd8d30b5abeababe6b70f660084b7dc16f
3
+ size 2060252928
qwen3.8-27b-hauhau-w8a8-vision-mtp/model-00009.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c9e95379f567007f312a5e24033bc59935b1ee4c2df7f6356a2f3e5ca434c204
3
+ size 2039276792
qwen3.8-27b-hauhau-w8a8-vision-mtp/model-00010.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3191b39e99871d97be9e1b6bc5485c3caa7e77ce54a5a4cde8269766461cf932
3
+ size 2086456832
qwen3.8-27b-hauhau-w8a8-vision-mtp/model-00011.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:cea09579aa996755bc365ab2ab0ae501998b48672b0b12fd607930e2fcf741e0
3
+ size 2013071632
qwen3.8-27b-hauhau-w8a8-vision-mtp/model-00012.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bc531dde00bd7609180aa18b282e391a48de1b3c3032016d647a5505cf1cb5d2
3
+ size 4334508936
qwen3.8-27b-hauhau-w8a8-vision-mtp/model-00013.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5a3db9777591c229af34aeb30198f99c7653b8d10beff38ffc752b6dae60b576
3
+ size 849400392
qwen3.8-27b-hauhau-w8a8-vision-mtp/model.safetensors.index.json ADDED
The diff for this file is too large to render. See raw diff
 
qwen3.8-27b-hauhau-w8a8-vision-mtp/processor_config.json ADDED
@@ -0,0 +1,60 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "image_processor": {
3
+ "do_convert_rgb": true,
4
+ "do_normalize": true,
5
+ "do_rescale": true,
6
+ "do_resize": true,
7
+ "image_mean": [
8
+ 0.5,
9
+ 0.5,
10
+ 0.5
11
+ ],
12
+ "image_processor_type": "Qwen2VLImageProcessor",
13
+ "image_std": [
14
+ 0.5,
15
+ 0.5,
16
+ 0.5
17
+ ],
18
+ "merge_size": 2,
19
+ "patch_size": 16,
20
+ "resample": 3,
21
+ "rescale_factor": 0.00392156862745098,
22
+ "size": {
23
+ "longest_edge": 16777216,
24
+ "shortest_edge": 65536
25
+ },
26
+ "temporal_patch_size": 2
27
+ },
28
+ "processor_class": "Qwen3VLProcessor",
29
+ "video_processor": {
30
+ "do_convert_rgb": true,
31
+ "do_normalize": true,
32
+ "do_rescale": true,
33
+ "do_resize": true,
34
+ "do_sample_frames": true,
35
+ "fps": 2,
36
+ "image_mean": [
37
+ 0.5,
38
+ 0.5,
39
+ 0.5
40
+ ],
41
+ "image_std": [
42
+ 0.5,
43
+ 0.5,
44
+ 0.5
45
+ ],
46
+ "max_frames": 768,
47
+ "merge_size": 2,
48
+ "min_frames": 4,
49
+ "patch_size": 16,
50
+ "resample": 3,
51
+ "rescale_factor": 0.00392156862745098,
52
+ "return_metadata": false,
53
+ "size": {
54
+ "longest_edge": 25165824,
55
+ "shortest_edge": 4096
56
+ },
57
+ "temporal_patch_size": 2,
58
+ "video_processor_type": "Qwen3VLVideoProcessor"
59
+ }
60
+ }
qwen3.8-27b-hauhau-w8a8-vision-mtp/results/precision-comparison.json ADDED
@@ -0,0 +1,382 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "date_utc": "2026-10-06",
3
+ "hardware": "2 x NVIDIA RTX 3090 24 GiB, NVLink, 125 GiB host RAM",
4
+ "vllm": "0.31.0",
5
+ "context_limit": 200000,
6
+ "kv_cache": "turboquant_4bit_nc",
7
+ "host_kv_offload_gib": 90,
8
+ "tensor_parallel_size": 2,
9
+ "sampling": {
10
+ "temperature": 0.6,
11
+ "top_p": 0.95,
12
+ "top_k": 20,
13
+ "presence_penalty": 0.0
14
+ },
15
+ "thinking": true,
16
+ "variants": {
17
+ "hauhau-w8a8": {
18
+ "long_context": {
19
+ "variant": "hauhau-w8a8",
20
+ "baseline": "model-tools/endy-long-context-20261005T172133Z",
21
+ "turns": 8,
22
+ "checks_passed": 53,
23
+ "checks_total": 54,
24
+ "strict_json_turns": 7,
25
+ "tool_calls": 2,
26
+ "median_decode_tokens_per_second": 46.885000000000005,
27
+ "grading_note": "Original checks retained; inspect semantic cyber findings separately."
28
+ },
29
+ "cybersecurity": {
30
+ "variant": "hauhau-w8a8",
31
+ "cases": 6,
32
+ "checks_passed": 30,
33
+ "checks_total": 43,
34
+ "strict_json_cases": 4,
35
+ "checks_passed_after_clarification": 34,
36
+ "note": "Small synthetic assessment, not a published benchmark or comprehensive security validation."
37
+ },
38
+ "speed": {
39
+ "variant": "hauhau-w8a8",
40
+ "output_tokens_per_request": 1024,
41
+ "sampling": {
42
+ "temperature": 0.6,
43
+ "top_p": 0.95,
44
+ "top_k": 20,
45
+ "presence_penalty": 0
46
+ },
47
+ "thinking_enabled": true,
48
+ "requests": [
49
+ {
50
+ "label": "short_0",
51
+ "input_tokens": 34,
52
+ "output_tokens": 1024,
53
+ "wall_seconds": 6.534,
54
+ "first_token_seconds": 0.753,
55
+ "decode_tokens_per_second": 177.144,
56
+ "warmup": true
57
+ },
58
+ {
59
+ "label": "short_1",
60
+ "input_tokens": 34,
61
+ "output_tokens": 1024,
62
+ "wall_seconds": 6.022,
63
+ "first_token_seconds": 0.126,
64
+ "decode_tokens_per_second": 173.698,
65
+ "warmup": false
66
+ },
67
+ {
68
+ "label": "short_2",
69
+ "input_tokens": 34,
70
+ "output_tokens": 1024,
71
+ "wall_seconds": 5.938,
72
+ "first_token_seconds": 0.116,
73
+ "decode_tokens_per_second": 175.89,
74
+ "warmup": false
75
+ },
76
+ {
77
+ "label": "long_0",
78
+ "input_tokens": 121616,
79
+ "output_tokens": 1024,
80
+ "wall_seconds": 40.625,
81
+ "first_token_seconds": 18.331,
82
+ "decode_tokens_per_second": 45.932,
83
+ "warmup": true
84
+ },
85
+ {
86
+ "label": "long_1",
87
+ "input_tokens": 121616,
88
+ "output_tokens": 1024,
89
+ "wall_seconds": 23.206,
90
+ "first_token_seconds": 1.08,
91
+ "decode_tokens_per_second": 46.28,
92
+ "warmup": false
93
+ },
94
+ {
95
+ "label": "long_2",
96
+ "input_tokens": 121616,
97
+ "output_tokens": 1024,
98
+ "wall_seconds": 23.118,
99
+ "first_token_seconds": 1.079,
100
+ "decode_tokens_per_second": 46.462,
101
+ "warmup": false
102
+ }
103
+ ],
104
+ "short_median_decode_tps": 174.79399999999998,
105
+ "long_median_decode_tps": 46.371,
106
+ "four_concurrent": {
107
+ "wall_seconds": 11.758,
108
+ "aggregate_output_tokens_per_second": 348.353,
109
+ "requests": [
110
+ {
111
+ "label": "concurrent_0",
112
+ "input_tokens": 34,
113
+ "output_tokens": 1024,
114
+ "wall_seconds": 11.74,
115
+ "first_token_seconds": 0.128,
116
+ "decode_tokens_per_second": 88.188
117
+ },
118
+ {
119
+ "label": "concurrent_1",
120
+ "input_tokens": 34,
121
+ "output_tokens": 1024,
122
+ "wall_seconds": 11.754,
123
+ "first_token_seconds": 3.292,
124
+ "decode_tokens_per_second": 121.005
125
+ },
126
+ {
127
+ "label": "concurrent_2",
128
+ "input_tokens": 34,
129
+ "output_tokens": 1024,
130
+ "wall_seconds": 11.753,
131
+ "first_token_seconds": 3.291,
132
+ "decode_tokens_per_second": 121.006
133
+ },
134
+ {
135
+ "label": "concurrent_3",
136
+ "input_tokens": 34,
137
+ "output_tokens": 1024,
138
+ "wall_seconds": 11.752,
139
+ "first_token_seconds": 3.29,
140
+ "decode_tokens_per_second": 121.009
141
+ }
142
+ ]
143
+ }
144
+ },
145
+ "custom_metadata": {
146
+ "valid_tool_cases": 6,
147
+ "total_cases": 6,
148
+ "correct_fields": 30,
149
+ "total_fields": 30,
150
+ "raw_json_cases": 5
151
+ },
152
+ "executed_sql_fixture": {
153
+ "passed": 2,
154
+ "total": 2
155
+ },
156
+ "native_api_integration": [
157
+ {
158
+ "test": "model_alias_and_underlying_checkpoint",
159
+ "passed": true
160
+ },
161
+ {
162
+ "test": "anthropic_text_through_caddy",
163
+ "passed": true
164
+ },
165
+ {
166
+ "test": "anthropic_streaming_through_caddy",
167
+ "passed": true
168
+ },
169
+ {
170
+ "test": "anthropic_tool_call_and_result_roundtrip",
171
+ "passed": true
172
+ },
173
+ {
174
+ "test": "anthropic_vision_two_distinct_images",
175
+ "passed": true
176
+ },
177
+ {
178
+ "test": "eight_concurrent_anthropic_requests",
179
+ "passed": true,
180
+ "wall_seconds": 1.0,
181
+ "total_output_tokens": 578
182
+ },
183
+ {
184
+ "test": "public_https_anthropic_endpoint",
185
+ "passed": true
186
+ }
187
+ ],
188
+ "claude_code": {
189
+ "wall_seconds": 2.721,
190
+ "exit_code": 0
191
+ }
192
+ },
193
+ "hauhau-w8a16": {
194
+ "long_context": {
195
+ "variant": "hauhau-w8a16",
196
+ "baseline": "/home/isnotreal/model-tools/endy-long-context-20261005T172133Z",
197
+ "turns": 8,
198
+ "checks_passed": 53,
199
+ "checks_total": 54,
200
+ "strict_json_turns": 7,
201
+ "tool_calls": 2,
202
+ "median_decode_tokens_per_second": 47.269999999999996,
203
+ "grading_note": "Original checks retained; inspect semantic cyber findings separately."
204
+ },
205
+ "cybersecurity": {
206
+ "variant": "hauhau-w8a16",
207
+ "cases": 6,
208
+ "checks_passed": 28,
209
+ "checks_total": 43,
210
+ "strict_json_cases": 6,
211
+ "checks_passed_after_clarification": 36,
212
+ "note": "Small synthetic assessment, not a published benchmark or comprehensive security validation."
213
+ },
214
+ "speed": {
215
+ "variant": "hauhau-w8a16",
216
+ "output_tokens_per_request": 1024,
217
+ "sampling": {
218
+ "temperature": 0.6,
219
+ "top_p": 0.95,
220
+ "top_k": 20,
221
+ "presence_penalty": 0
222
+ },
223
+ "thinking_enabled": true,
224
+ "requests": [
225
+ {
226
+ "label": "short_0",
227
+ "input_tokens": 34,
228
+ "output_tokens": 1024,
229
+ "wall_seconds": 5.891,
230
+ "first_token_seconds": 0.131,
231
+ "decode_tokens_per_second": 177.768,
232
+ "warmup": true
233
+ },
234
+ {
235
+ "label": "short_1",
236
+ "input_tokens": 34,
237
+ "output_tokens": 1024,
238
+ "wall_seconds": 6.107,
239
+ "first_token_seconds": 0.113,
240
+ "decode_tokens_per_second": 170.834,
241
+ "warmup": false
242
+ },
243
+ {
244
+ "label": "short_2",
245
+ "input_tokens": 34,
246
+ "output_tokens": 1024,
247
+ "wall_seconds": 6.021,
248
+ "first_token_seconds": 0.115,
249
+ "decode_tokens_per_second": 173.38,
250
+ "warmup": false
251
+ },
252
+ {
253
+ "label": "long_0",
254
+ "input_tokens": 121616,
255
+ "output_tokens": 1024,
256
+ "wall_seconds": 44.096,
257
+ "first_token_seconds": 21.895,
258
+ "decode_tokens_per_second": 46.124,
259
+ "warmup": true
260
+ },
261
+ {
262
+ "label": "long_1",
263
+ "input_tokens": 121616,
264
+ "output_tokens": 1024,
265
+ "wall_seconds": 23.291,
266
+ "first_token_seconds": 1.18,
267
+ "decode_tokens_per_second": 46.312,
268
+ "warmup": false
269
+ },
270
+ {
271
+ "label": "long_2",
272
+ "input_tokens": 121616,
273
+ "output_tokens": 1024,
274
+ "wall_seconds": 23.12,
275
+ "first_token_seconds": 1.17,
276
+ "decode_tokens_per_second": 46.652,
277
+ "warmup": false
278
+ }
279
+ ],
280
+ "short_median_decode_tps": 172.107,
281
+ "long_median_decode_tps": 46.482,
282
+ "four_concurrent": {
283
+ "wall_seconds": 10.924,
284
+ "aggregate_output_tokens_per_second": 374.959,
285
+ "requests": [
286
+ {
287
+ "label": "concurrent_0",
288
+ "input_tokens": 34,
289
+ "output_tokens": 1024,
290
+ "wall_seconds": 10.907,
291
+ "first_token_seconds": 0.118,
292
+ "decode_tokens_per_second": 94.911
293
+ },
294
+ {
295
+ "label": "concurrent_1",
296
+ "input_tokens": 34,
297
+ "output_tokens": 1024,
298
+ "wall_seconds": 10.914,
299
+ "first_token_seconds": 0.221,
300
+ "decode_tokens_per_second": 95.758
301
+ },
302
+ {
303
+ "label": "concurrent_2",
304
+ "input_tokens": 34,
305
+ "output_tokens": 1024,
306
+ "wall_seconds": 10.921,
307
+ "first_token_seconds": 2.561,
308
+ "decode_tokens_per_second": 122.487
309
+ },
310
+ {
311
+ "label": "concurrent_3",
312
+ "input_tokens": 34,
313
+ "output_tokens": 1024,
314
+ "wall_seconds": 10.92,
315
+ "first_token_seconds": 2.56,
316
+ "decode_tokens_per_second": 122.494
317
+ }
318
+ ]
319
+ }
320
+ },
321
+ "custom_metadata": {
322
+ "valid_tool_cases": 6,
323
+ "total_cases": 6,
324
+ "correct_fields": 30,
325
+ "total_fields": 30,
326
+ "raw_json_cases": 4
327
+ },
328
+ "executed_sql_fixture": {
329
+ "passed": 2,
330
+ "total": 2
331
+ }
332
+ },
333
+ "endy-w8a16": {
334
+ "long_context": {
335
+ "variant": "endy-w8a16-vllm-0.31-qwen-coding-preset",
336
+ "baseline": "/home/isnotreal/model-tools/endy-long-context-20261005T172133Z",
337
+ "turns": 8,
338
+ "checks_passed": 46,
339
+ "checks_total": 54,
340
+ "strict_json_turns": 0,
341
+ "tool_calls": 0,
342
+ "median_decode_tokens_per_second": 47.545,
343
+ "grading_note": "Original checks retained; inspect semantic cyber findings separately."
344
+ },
345
+ "custom_metadata": {
346
+ "fresh_valid_tool_cases": 2,
347
+ "fresh_cases": 2,
348
+ "compacted_valid_tool_cases": 0,
349
+ "compacted_cases": 2,
350
+ "long_valid_tool_cases": 0,
351
+ "long_cases": 2
352
+ },
353
+ "note": "Endy can call tools on fresh prompts. Reliability degraded on the compacted and long-history custom-metadata fixtures."
354
+ },
355
+ "endy-w8a8": {
356
+ "long_context": {
357
+ "variant": "endy-w8a8-vllm-0.31",
358
+ "baseline": "/home/isnotreal/model-tools/endy-long-context-20261005T172133Z",
359
+ "turns": 8,
360
+ "checks_passed": 27,
361
+ "checks_total": 54,
362
+ "strict_json_turns": 0,
363
+ "tool_calls": 0,
364
+ "median_decode_tokens_per_second": 47.515,
365
+ "grading_note": "Original checks retained; inspect semantic cyber findings separately."
366
+ },
367
+ "note": "Earlier W8A8 long-context run used different sampling; not a matched activation-precision comparison."
368
+ }
369
+ },
370
+ "limitations": [
371
+ "Small synthetic local assessment, not a published cybersecurity benchmark.",
372
+ "Long-context tests independently replayed eight original 122k-124k-token prefixes, not an eight-turn fresh conversation.",
373
+ "The one failed long-context strict check per Hauhau variant was a literal regex mismatch; the proposed SQL was executed successfully in a fixture.",
374
+ "The four-concurrent speed measurement is one batch and can include first-shape compilation; it is not a stable concurrency ranking.",
375
+ "Both variants made substantive security-policy mistakes on the broader quiz; manual review is necessary.",
376
+ "W8A8 SSRF clarification exhausted the 4096-output-token cap and truncated JSON; its secondary score includes that failure.",
377
+ "The native W8A8 export uses dynamic INT8 activations throughout eligible language linears; the older SmoothQuant export is a different mixed W8A8/W8A16 artifact.",
378
+ "Hauhau weights originate in Q8_K_P GGUF, then reconstruction and INT8 conversion. W8A16 and W8A8 use identical INT8 weight values, but native kernels and scale handling differ.",
379
+ "MTP and vision weights are retained; speculative decoding was not enabled.",
380
+ "Two serial repeats after warmup and one concurrency batch are too few to attribute small differences to precision alone."
381
+ ]
382
+ }
qwen3.8-27b-hauhau-w8a8-vision-mtp/tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:06b9509352d2af50381ab2247e083b80d32d5c0aba91c272ca9ff729b6a0e523
3
+ size 19989325
qwen3.8-27b-hauhau-w8a8-vision-mtp/tokenizer_config.json ADDED
@@ -0,0 +1,33 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "audio_bos_token": "<|audio_start|>",
4
+ "audio_eos_token": "<|audio_end|>",
5
+ "audio_token": "<|audio_pad|>",
6
+ "backend": "tokenizers",
7
+ "bos_token": null,
8
+ "clean_up_tokenization_spaces": false,
9
+ "eos_token": "<|im_end|>",
10
+ "errors": "replace",
11
+ "image_token": "<|image_pad|>",
12
+ "is_local": true,
13
+ "local_files_only": false,
14
+ "model_max_length": 262144,
15
+ "model_specific_special_tokens": {
16
+ "audio_bos_token": "<|audio_start|>",
17
+ "audio_eos_token": "<|audio_end|>",
18
+ "audio_token": "<|audio_pad|>",
19
+ "image_token": "<|image_pad|>",
20
+ "video_token": "<|video_pad|>",
21
+ "vision_bos_token": "<|vision_start|>",
22
+ "vision_eos_token": "<|vision_end|>"
23
+ },
24
+ "pad_token": "<|endoftext|>",
25
+ "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
26
+ "processor_class": "Qwen3VLProcessor",
27
+ "split_special_tokens": false,
28
+ "tokenizer_class": "Qwen2Tokenizer",
29
+ "unk_token": null,
30
+ "video_token": "<|video_pad|>",
31
+ "vision_bos_token": "<|vision_start|>",
32
+ "vision_eos_token": "<|vision_end|>"
33
+ }
qwen3.8-27b-hauhau-w8a8-vision-mtp/vision.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d7defc90994f7bbc58a8feb8d3c4144895babb434a9be9f47e88ae3646e1df33
3
+ size 921497224