Add files using upload-large-folder tool
Browse files- README.md +18 -15
- chat_template.jinja +45 -126
- tokenizer_config.json +4 -4
README.md
CHANGED
|
@@ -1,20 +1,21 @@
|
|
| 1 |
---
|
| 2 |
base_model: TeichAI/gemma-4-31B-it-Claude-Opus-Distill
|
|
|
|
|
|
|
|
|
|
| 3 |
tags:
|
| 4 |
-
-
|
| 5 |
-
-
|
| 6 |
-
-
|
| 7 |
-
-
|
| 8 |
-
-
|
| 9 |
-
-
|
| 10 |
license: apache-2.0
|
| 11 |
datasets:
|
| 12 |
-
- TeichAI/Claude-Opus-4.6-Reasoning-887x
|
| 13 |
-
- TeichAI/Claude-Sonnet-4.6-Reasoning-1100x
|
| 14 |
-
- TeichAI/claude-4.5-opus-high-reasoning-250x
|
| 15 |
-
- Crownelius/Opus-4.6-Reasoning-2100x-formatted
|
| 16 |
-
library_name: mlx
|
| 17 |
-
pipeline_tag: image-text-to-text
|
| 18 |
---
|
| 19 |
# 🦆 zecanard/gemma-4-31B-it-Claude-Opus-Distilled-MLX-8bit-mxfp8
|
| 20 |
|
|
@@ -23,9 +24,9 @@ Please refer to the [original model card](https://huggingface.co/TeichAI/gemma-4
|
|
| 23 |
|
| 24 |
## 🌟 Quality
|
| 25 |
|
| 26 |
-
Quantized vision language model with **8.564 bits per weight**.
|
| 27 |
|
| 28 |
-
`mlx_vlm.convert --
|
| 29 |
|
| 30 |
## 🛠️ Customizations
|
| 31 |
|
|
@@ -33,6 +34,8 @@ This quant is aware of the current date, and also enables thinking (if available
|
|
| 33 |
|
| 34 |
`{%- set enable_thinking = true %}`
|
| 35 |
|
|
|
|
|
|
|
| 36 |
## 🖥️ Use with `mlx`
|
| 37 |
|
| 38 |
```bash
|
|
@@ -40,5 +43,5 @@ pip install -U mlx-vlm
|
|
| 40 |
```
|
| 41 |
|
| 42 |
```bash
|
| 43 |
-
mlx_vlm.generate --model zecanard/gemma-4-31B-it-Claude-Opus-Distilled-MLX-8bit-mxfp8 --max-tokens 100 --temperature 0
|
| 44 |
```
|
|
|
|
| 1 |
---
|
| 2 |
base_model: TeichAI/gemma-4-31B-it-Claude-Opus-Distill
|
| 3 |
+
language: en
|
| 4 |
+
pipeline_tag: image-text-to-text
|
| 5 |
+
library_name: mlx
|
| 6 |
tags:
|
| 7 |
+
- mlx
|
| 8 |
+
- text-generation-inference
|
| 9 |
+
- transformers
|
| 10 |
+
- unsloth
|
| 11 |
+
- gemma4
|
| 12 |
+
- reasoning
|
| 13 |
license: apache-2.0
|
| 14 |
datasets:
|
| 15 |
+
- TeichAI/Claude-Opus-4.6-Reasoning-887x
|
| 16 |
+
- TeichAI/Claude-Sonnet-4.6-Reasoning-1100x
|
| 17 |
+
- TeichAI/claude-4.5-opus-high-reasoning-250x
|
| 18 |
+
- Crownelius/Opus-4.6-Reasoning-2100x-formatted
|
|
|
|
|
|
|
| 19 |
---
|
| 20 |
# 🦆 zecanard/gemma-4-31B-it-Claude-Opus-Distilled-MLX-8bit-mxfp8
|
| 21 |
|
|
|
|
| 24 |
|
| 25 |
## 🌟 Quality
|
| 26 |
|
| 27 |
+
Quantized vision language model with an effective **8.564 bits per weight**.
|
| 28 |
|
| 29 |
+
`mlx_vlm.convert --quantize --q-bits 8 --q-group-size 32 --q-mode mxfp8`
|
| 30 |
|
| 31 |
## 🛠️ Customizations
|
| 32 |
|
|
|
|
| 34 |
|
| 35 |
`{%- set enable_thinking = true %}`
|
| 36 |
|
| 37 |
+
You may also need to adjust your environment’s **Reasoning Section Parsing** to recognize `<|channel>thought` as the **Start String**, and `<channel|>` as the **End String**.
|
| 38 |
+
|
| 39 |
## 🖥️ Use with `mlx`
|
| 40 |
|
| 41 |
```bash
|
|
|
|
| 43 |
```
|
| 44 |
|
| 45 |
```bash
|
| 46 |
+
mlx_vlm.generate --model zecanard/gemma-4-31B-it-Claude-Opus-Distilled-MLX-8bit-mxfp8 --max-tokens 100 --temperature 0 --prompt "Describe this image." --image <path_to_image>
|
| 47 |
```
|
chat_template.jinja
CHANGED
|
@@ -3,7 +3,7 @@
|
|
| 3 |
{%- if strftime_now is defined %}
|
| 4 |
{%- set currentDate = strftime_now("%Y-%m-%d %G:%i:%s") %}
|
| 5 |
{%- elif not currentDate is defined %}
|
| 6 |
-
{%- set currentDate = "2026-04-
|
| 7 |
{%- endif %}
|
| 8 |
{{- "The current date is: " + currentDate + ".\n\n" }}
|
| 9 |
{#- Template customizations by @zecanard. #}
|
|
@@ -20,15 +20,34 @@
|
|
| 20 |
description:<|"|>{{ value['description'] }}<|"|>
|
| 21 |
{%- set add_comma = true -%}
|
| 22 |
{%- endif -%}
|
|
|
|
|
|
|
|
|
|
|
|
|
| 23 |
{%- if value['type'] | upper == 'STRING' -%}
|
| 24 |
{%- if value['enum'] -%}
|
| 25 |
{%- if add_comma %},{%- else -%} {%- set add_comma = true -%} {% endif -%}
|
| 26 |
enum:{{ format_argument(value['enum']) }}
|
| 27 |
{%- endif -%}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 28 |
{%- elif value['type'] | upper == 'ARRAY' -%}
|
| 29 |
{%- if value['items'] is mapping and value['items'] -%}
|
| 30 |
-
|
| 31 |
-
items:{
|
| 32 |
{%- set ns_items = namespace(found_first=false) -%}
|
| 33 |
{%- for item_key, item_value in value['items'] | dictsort -%}
|
| 34 |
{%- if item_value is not none -%}
|
|
@@ -61,32 +80,6 @@
|
|
| 61 |
}
|
| 62 |
{%- endif -%}
|
| 63 |
{%- endif -%}
|
| 64 |
-
{%- if value['nullable'] %}
|
| 65 |
-
{%- if add_comma %},{%- else -%} {%- set add_comma = true -%} {% endif -%}
|
| 66 |
-
nullable:true
|
| 67 |
-
{%- endif -%}
|
| 68 |
-
{%- if value['type'] | upper == 'OBJECT' -%}
|
| 69 |
-
{%- if value['properties'] is defined and value['properties'] is mapping -%}
|
| 70 |
-
{%- if add_comma %},{%- else -%} {%- set add_comma = true -%} {% endif -%}
|
| 71 |
-
properties:{
|
| 72 |
-
{{- format_parameters(value['properties'], value['required'] | default([])) -}}
|
| 73 |
-
}
|
| 74 |
-
{%- elif value is mapping -%}
|
| 75 |
-
{%- if add_comma %},{%- else -%} {%- set add_comma = true -%} {% endif -%}
|
| 76 |
-
properties:{
|
| 77 |
-
{{- format_parameters(value, value['required'] | default([])) -}}
|
| 78 |
-
}
|
| 79 |
-
{%- endif -%}
|
| 80 |
-
{%- if value['required'] -%}
|
| 81 |
-
{%- if add_comma %},{%- else -%} {%- set add_comma = true -%} {% endif -%}
|
| 82 |
-
required:[
|
| 83 |
-
{%- for item in value['required'] | default([]) -%}
|
| 84 |
-
<|"|>{{- item -}}<|"|>
|
| 85 |
-
{%- if not loop.last %},{% endif -%}
|
| 86 |
-
{%- endfor -%}
|
| 87 |
-
]
|
| 88 |
-
{%- endif -%}
|
| 89 |
-
{%- endif -%}
|
| 90 |
{%- if add_comma %},{%- else -%} {%- set add_comma = true -%} {% endif -%}
|
| 91 |
type:<|"|>{{ value['type'] | upper }}<|"|>}
|
| 92 |
{%- endif -%}
|
|
@@ -166,31 +159,16 @@
|
|
| 166 |
{{- ns.result | trim -}}
|
| 167 |
{%- endmacro -%}
|
| 168 |
|
| 169 |
-
{%- macro format_tool_response_block(tool_name, response) -%}
|
| 170 |
-
{{- '<|tool_response>' -}}
|
| 171 |
-
{%- if response is mapping -%}
|
| 172 |
-
{{- 'response:' + tool_name + '{' -}}
|
| 173 |
-
{%- for key, value in response | dictsort -%}
|
| 174 |
-
{{- key -}}:{{- format_argument(value, escape_keys=False) -}}
|
| 175 |
-
{%- if not loop.last %},{% endif -%}
|
| 176 |
-
{%- endfor -%}
|
| 177 |
-
{{- '}' -}}
|
| 178 |
-
{%- else -%}
|
| 179 |
-
{{- 'response:' + tool_name + '{value:' + format_argument(response, escape_keys=False) + '}' -}}
|
| 180 |
-
{%- endif -%}
|
| 181 |
-
{{- '<tool_response|>' -}}
|
| 182 |
-
{%- endmacro -%}
|
| 183 |
-
|
| 184 |
{%- set ns = namespace(prev_message_type=None) -%}
|
| 185 |
{%- set loop_messages = messages -%}
|
| 186 |
-
{{
|
| 187 |
{#- Handle System/Tool Definitions Block -#}
|
| 188 |
{%- if enable_thinking | default(true) or tools or messages[0]['role'] in ['system', 'developer'] -%}
|
| 189 |
{{- '<|turn>system\n' -}}
|
| 190 |
|
| 191 |
{#- Inject Thinking token at the very top of the FIRST system turn -#}
|
| 192 |
{%- if enable_thinking | default(true) -%}
|
| 193 |
-
{{- '<|think|>
|
| 194 |
{%- set ns.prev_message_type = 'think' -%}
|
| 195 |
{%- endif -%}
|
| 196 |
|
|
@@ -211,41 +189,11 @@
|
|
| 211 |
{{- '<turn|>\n' -}}
|
| 212 |
{%- endif %}
|
| 213 |
|
| 214 |
-
{#- Pre-scan: find last user message index for reasoning guard -#}
|
| 215 |
-
{%- set ns_turn = namespace(last_user_idx=-1) -%}
|
| 216 |
-
{%- for i in range(loop_messages | length) -%}
|
| 217 |
-
{%- if loop_messages[i]['role'] == 'user' -%}
|
| 218 |
-
{%- set ns_turn.last_user_idx = i -%}
|
| 219 |
-
{%- endif -%}
|
| 220 |
-
{%- endfor -%}
|
| 221 |
-
|
| 222 |
{#- Loop through messages -#}
|
| 223 |
{%- for message in loop_messages -%}
|
| 224 |
-
{%- if message['role'] != 'tool' -%}
|
| 225 |
{%- set ns.prev_message_type = None -%}
|
| 226 |
{%- set role = 'model' if message['role'] == 'assistant' else message['role'] -%}
|
| 227 |
-
{#- Detect continuation: suppress duplicate <|turn>model when previous non-tool message was also assistant -#}
|
| 228 |
-
{%- set prev_nt = namespace(role=None, found=false) -%}
|
| 229 |
-
{%- if loop.index0 > 0 -%}
|
| 230 |
-
{%- for j in range(loop.index0 - 1, -1, -1) -%}
|
| 231 |
-
{%- if not prev_nt.found -%}
|
| 232 |
-
{%- if loop_messages[j]['role'] != 'tool' -%}
|
| 233 |
-
{%- set prev_nt.role = loop_messages[j]['role'] -%}
|
| 234 |
-
{%- set prev_nt.found = true -%}
|
| 235 |
-
{%- endif -%}
|
| 236 |
-
{%- endif -%}
|
| 237 |
-
{%- endfor -%}
|
| 238 |
-
{%- endif -%}
|
| 239 |
-
{%- set continue_same_model_turn = (role == 'model' and prev_nt.role == 'assistant') -%}
|
| 240 |
-
{%- if not continue_same_model_turn -%}
|
| 241 |
{{- '<|turn>' + role + '\n' }}
|
| 242 |
-
{%- endif -%}
|
| 243 |
-
|
| 244 |
-
{#- Render reasoning/reasoning_content as thinking channel -#}
|
| 245 |
-
{%- set thinking_text = message.get('reasoning') or message.get('reasoning_content') -%}
|
| 246 |
-
{%- if thinking_text and loop.index0 > ns_turn.last_user_idx -%}
|
| 247 |
-
{{- '<|channel>thought\n' + thinking_text + '\n<channel|>' -}}
|
| 248 |
-
{%- endif -%}
|
| 249 |
|
| 250 |
{%- if message['tool_calls'] -%}
|
| 251 |
{%- for tool_call in message['tool_calls'] -%}
|
|
@@ -266,49 +214,23 @@
|
|
| 266 |
{%- set ns.prev_message_type = 'tool_call' -%}
|
| 267 |
{%- endif -%}
|
| 268 |
|
| 269 |
-
{%-
|
| 270 |
-
|
| 271 |
-
{#- Legacy: tool_responses embedded on the assistant message (Google/Gemma native) -#}
|
| 272 |
{%- for tool_response in message['tool_responses'] -%}
|
| 273 |
-
{{-
|
| 274 |
-
{%-
|
| 275 |
-
|
| 276 |
-
|
| 277 |
-
|
| 278 |
-
|
| 279 |
-
{%- set ns_tool_scan = namespace(stopped=false) -%}
|
| 280 |
-
{%- for k in range(loop.index0 + 1, loop_messages | length) -%}
|
| 281 |
-
{%- if ns_tool_scan.stopped -%}
|
| 282 |
-
{%- elif loop_messages[k]['role'] != 'tool' -%}
|
| 283 |
-
{%- set ns_tool_scan.stopped = true -%}
|
| 284 |
-
{%- else -%}
|
| 285 |
-
{%- set follow = loop_messages[k] -%}
|
| 286 |
-
{#- Resolve tool_call_id to function name -#}
|
| 287 |
-
{%- set ns_tname = namespace(name=follow.get('name') | default('unknown')) -%}
|
| 288 |
-
{%- for tc in message['tool_calls'] -%}
|
| 289 |
-
{%- if tc.get('id') == follow.get('tool_call_id') -%}
|
| 290 |
-
{%- set ns_tname.name = tc['function']['name'] -%}
|
| 291 |
-
{%- endif -%}
|
| 292 |
{%- endfor -%}
|
| 293 |
-
{
|
| 294 |
-
|
| 295 |
-
{
|
| 296 |
-
{{- format_tool_response_block(ns_tname.name, tool_body) -}}
|
| 297 |
-
{%- elif tool_body is sequence and tool_body is not string -%}
|
| 298 |
-
{%- set ns_txt = namespace(s='') -%}
|
| 299 |
-
{%- for part in tool_body -%}
|
| 300 |
-
{%- if part.get('type') == 'text' -%}
|
| 301 |
-
{%- set ns_txt.s = ns_txt.s + (part.get('text') | default('')) -%}
|
| 302 |
-
{%- endif -%}
|
| 303 |
-
{%- endfor -%}
|
| 304 |
-
{{- format_tool_response_block(ns_tname.name, ns_txt.s) -}}
|
| 305 |
-
{%- else -%}
|
| 306 |
-
{{- format_tool_response_block(ns_tname.name, tool_body) -}}
|
| 307 |
-
{%- endif -%}
|
| 308 |
-
{%- set ns_tr_out.flag = true -%}
|
| 309 |
-
{%- set ns.prev_message_type = 'tool_response' -%}
|
| 310 |
{%- endif -%}
|
|
|
|
| 311 |
{%- endfor -%}
|
|
|
|
| 312 |
{%- endif -%}
|
| 313 |
|
| 314 |
{%- if message['content'] is string -%}
|
|
@@ -326,31 +248,28 @@
|
|
| 326 |
{{- item['text'] | trim -}}
|
| 327 |
{%- endif -%}
|
| 328 |
{%- elif item['type'] == 'image' -%}
|
| 329 |
-
{{- '<|image|>' -}}
|
| 330 |
{%- set ns.prev_message_type = 'image' -%}
|
| 331 |
{%- elif item['type'] == 'audio' -%}
|
| 332 |
{{- '<|audio|>' -}}
|
| 333 |
{%- set ns.prev_message_type = 'audio' -%}
|
| 334 |
{%- elif item['type'] == 'video' -%}
|
| 335 |
-
{{- '<|video|>' -}}
|
| 336 |
{%- set ns.prev_message_type = 'video' -%}
|
| 337 |
{%- endif -%}
|
| 338 |
{%- endfor -%}
|
| 339 |
{%- endif -%}
|
| 340 |
|
| 341 |
-
{%- if
|
| 342 |
-
{{- '<|tool_response>' -}}
|
| 343 |
-
{%- elif not (ns_tr_out.flag and not message.get('content')) -%}
|
| 344 |
{{- '<turn|>\n' -}}
|
| 345 |
{%- endif -%}
|
| 346 |
-
{%- endif -%}
|
| 347 |
{%- endfor -%}
|
| 348 |
|
| 349 |
{%- if add_generation_prompt -%}
|
| 350 |
-
{%- if ns.prev_message_type != 'tool_response'
|
| 351 |
{{- '<|turn>model\n' -}}
|
| 352 |
-
{%- if not enable_thinking | default(true) -%}
|
| 353 |
-
{{- '<|channel>thought\n<channel|>' -}}
|
| 354 |
-
{%- endif -%}
|
| 355 |
{%- endif -%}
|
| 356 |
-
{%-
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
{%- if strftime_now is defined %}
|
| 4 |
{%- set currentDate = strftime_now("%Y-%m-%d %G:%i:%s") %}
|
| 5 |
{%- elif not currentDate is defined %}
|
| 6 |
+
{%- set currentDate = "2026-04-15" %}
|
| 7 |
{%- endif %}
|
| 8 |
{{- "The current date is: " + currentDate + ".\n\n" }}
|
| 9 |
{#- Template customizations by @zecanard. #}
|
|
|
|
| 20 |
description:<|"|>{{ value['description'] }}<|"|>
|
| 21 |
{%- set add_comma = true -%}
|
| 22 |
{%- endif -%}
|
| 23 |
+
{%- if value['nullable'] %}
|
| 24 |
+
{%- if add_comma %},{%- else -%} {%- set add_comma = true -%} {% endif -%}
|
| 25 |
+
nullable:true
|
| 26 |
+
{%- endif -%}
|
| 27 |
{%- if value['type'] | upper == 'STRING' -%}
|
| 28 |
{%- if value['enum'] -%}
|
| 29 |
{%- if add_comma %},{%- else -%} {%- set add_comma = true -%} {% endif -%}
|
| 30 |
enum:{{ format_argument(value['enum']) }}
|
| 31 |
{%- endif -%}
|
| 32 |
+
{%- elif value['type'] | upper == 'OBJECT' -%}
|
| 33 |
+
,properties:{
|
| 34 |
+
{%- if value['properties'] is defined and value['properties'] is mapping -%}
|
| 35 |
+
{{- format_parameters(value['properties'], value['required'] | default([])) -}}
|
| 36 |
+
{%- elif value is mapping -%}
|
| 37 |
+
{{- format_parameters(value, value['required'] | default([])) -}}
|
| 38 |
+
{%- endif -%}
|
| 39 |
+
}
|
| 40 |
+
{%- if value['required'] -%}
|
| 41 |
+
,required:[
|
| 42 |
+
{%- for item in value['required'] | default([]) -%}
|
| 43 |
+
<|"|>{{- item -}}<|"|>
|
| 44 |
+
{%- if not loop.last %},{% endif -%}
|
| 45 |
+
{%- endfor -%}
|
| 46 |
+
]
|
| 47 |
+
{%- endif -%}
|
| 48 |
{%- elif value['type'] | upper == 'ARRAY' -%}
|
| 49 |
{%- if value['items'] is mapping and value['items'] -%}
|
| 50 |
+
,items:{
|
|
|
|
| 51 |
{%- set ns_items = namespace(found_first=false) -%}
|
| 52 |
{%- for item_key, item_value in value['items'] | dictsort -%}
|
| 53 |
{%- if item_value is not none -%}
|
|
|
|
| 80 |
}
|
| 81 |
{%- endif -%}
|
| 82 |
{%- endif -%}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 83 |
{%- if add_comma %},{%- else -%} {%- set add_comma = true -%} {% endif -%}
|
| 84 |
type:<|"|>{{ value['type'] | upper }}<|"|>}
|
| 85 |
{%- endif -%}
|
|
|
|
| 159 |
{{- ns.result | trim -}}
|
| 160 |
{%- endmacro -%}
|
| 161 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 162 |
{%- set ns = namespace(prev_message_type=None) -%}
|
| 163 |
{%- set loop_messages = messages -%}
|
| 164 |
+
{{ bos_token }}
|
| 165 |
{#- Handle System/Tool Definitions Block -#}
|
| 166 |
{%- if enable_thinking | default(true) or tools or messages[0]['role'] in ['system', 'developer'] -%}
|
| 167 |
{{- '<|turn>system\n' -}}
|
| 168 |
|
| 169 |
{#- Inject Thinking token at the very top of the FIRST system turn -#}
|
| 170 |
{%- if enable_thinking | default(true) -%}
|
| 171 |
+
{{- '<|think|>' -}}
|
| 172 |
{%- set ns.prev_message_type = 'think' -%}
|
| 173 |
{%- endif -%}
|
| 174 |
|
|
|
|
| 189 |
{{- '<turn|>\n' -}}
|
| 190 |
{%- endif %}
|
| 191 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 192 |
{#- Loop through messages -#}
|
| 193 |
{%- for message in loop_messages -%}
|
|
|
|
| 194 |
{%- set ns.prev_message_type = None -%}
|
| 195 |
{%- set role = 'model' if message['role'] == 'assistant' else message['role'] -%}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 196 |
{{- '<|turn>' + role + '\n' }}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 197 |
|
| 198 |
{%- if message['tool_calls'] -%}
|
| 199 |
{%- for tool_call in message['tool_calls'] -%}
|
|
|
|
| 214 |
{%- set ns.prev_message_type = 'tool_call' -%}
|
| 215 |
{%- endif -%}
|
| 216 |
|
| 217 |
+
{%- if message['tool_responses'] -%}
|
| 218 |
+
{#- Tool Response handling -#}
|
|
|
|
| 219 |
{%- for tool_response in message['tool_responses'] -%}
|
| 220 |
+
{{- '<|tool_response>' -}}
|
| 221 |
+
{%- if tool_response['response'] is mapping -%}
|
| 222 |
+
{{- 'response:' + tool_response['name'] | default('unknown') + '{' -}}
|
| 223 |
+
{%- for key, value in tool_response['response'] | dictsort -%}
|
| 224 |
+
{{- key -}}:{{- format_argument(value, escape_keys=False) -}}
|
| 225 |
+
{%- if not loop.last %},{% endif -%}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 226 |
{%- endfor -%}
|
| 227 |
+
{{- '}' -}}
|
| 228 |
+
{%- else -%}
|
| 229 |
+
{{- 'response:' + tool_response['name'] | default('unknown') + '{value:' + format_argument(tool_response['response'], escape_keys=False) + '}' -}}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 230 |
{%- endif -%}
|
| 231 |
+
{{- '<tool_response|>' -}}
|
| 232 |
{%- endfor -%}
|
| 233 |
+
{%- set ns.prev_message_type = 'tool_response' -%}
|
| 234 |
{%- endif -%}
|
| 235 |
|
| 236 |
{%- if message['content'] is string -%}
|
|
|
|
| 248 |
{{- item['text'] | trim -}}
|
| 249 |
{%- endif -%}
|
| 250 |
{%- elif item['type'] == 'image' -%}
|
| 251 |
+
{{- '\n\n<|image|>\n\n' -}}
|
| 252 |
{%- set ns.prev_message_type = 'image' -%}
|
| 253 |
{%- elif item['type'] == 'audio' -%}
|
| 254 |
{{- '<|audio|>' -}}
|
| 255 |
{%- set ns.prev_message_type = 'audio' -%}
|
| 256 |
{%- elif item['type'] == 'video' -%}
|
| 257 |
+
{{- '\n\n<|video|>\n\n' -}}
|
| 258 |
{%- set ns.prev_message_type = 'video' -%}
|
| 259 |
{%- endif -%}
|
| 260 |
{%- endfor -%}
|
| 261 |
{%- endif -%}
|
| 262 |
|
| 263 |
+
{%- if not (message['tool_responses'] and not message['content']) -%}
|
|
|
|
|
|
|
| 264 |
{{- '<turn|>\n' -}}
|
| 265 |
{%- endif -%}
|
|
|
|
| 266 |
{%- endfor -%}
|
| 267 |
|
| 268 |
{%- if add_generation_prompt -%}
|
| 269 |
+
{%- if ns.prev_message_type != 'tool_response' -%}
|
| 270 |
{{- '<|turn>model\n' -}}
|
|
|
|
|
|
|
|
|
|
| 271 |
{%- endif -%}
|
| 272 |
+
{%- if not enable_thinking | default(true) -%}
|
| 273 |
+
{{- '<|channel>thought\n<channel|>' -}}
|
| 274 |
+
{%- endif -%}
|
| 275 |
+
{%- endif -%}
|
tokenizer_config.json
CHANGED
|
@@ -7,7 +7,7 @@
|
|
| 7 |
"eoa_token": "<audio|>",
|
| 8 |
"eoc_token": "<channel|>",
|
| 9 |
"eoi_token": "<image|>",
|
| 10 |
-
"eos_token": "<
|
| 11 |
"eot_token": "<turn|>",
|
| 12 |
"escape_token": "<|\"|>",
|
| 13 |
"etc_token": "<tool_call|>",
|
|
@@ -19,7 +19,7 @@
|
|
| 19 |
"image_token": "<|image|>",
|
| 20 |
"is_local": true,
|
| 21 |
"mask_token": "<mask>",
|
| 22 |
-
"model_max_length":
|
| 23 |
"model_specific_special_tokens": {
|
| 24 |
"audio_token": "<|audio|>",
|
| 25 |
"boa_token": "<|audio>",
|
|
@@ -41,7 +41,7 @@
|
|
| 41 |
"think_token": "<|think|>"
|
| 42 |
},
|
| 43 |
"pad_token": "<pad>",
|
| 44 |
-
"padding_side": "
|
| 45 |
"processor_class": "Gemma4Processor",
|
| 46 |
"response_schema": {
|
| 47 |
"properties": {
|
|
@@ -82,7 +82,7 @@
|
|
| 82 |
}
|
| 83 |
},
|
| 84 |
"type": "object",
|
| 85 |
-
"x-regex": "(\\<\\|channel\\>thought\\n(?P<thinking>.*?)\\<channel\\|\\>)?(?P<
|
| 86 |
},
|
| 87 |
"soc_token": "<|channel>",
|
| 88 |
"sot_token": "<|turn>",
|
|
|
|
| 7 |
"eoa_token": "<audio|>",
|
| 8 |
"eoc_token": "<channel|>",
|
| 9 |
"eoi_token": "<image|>",
|
| 10 |
+
"eos_token": "<eos>",
|
| 11 |
"eot_token": "<turn|>",
|
| 12 |
"escape_token": "<|\"|>",
|
| 13 |
"etc_token": "<tool_call|>",
|
|
|
|
| 19 |
"image_token": "<|image|>",
|
| 20 |
"is_local": true,
|
| 21 |
"mask_token": "<mask>",
|
| 22 |
+
"model_max_length": 1000000000000000019884624838656,
|
| 23 |
"model_specific_special_tokens": {
|
| 24 |
"audio_token": "<|audio|>",
|
| 25 |
"boa_token": "<|audio>",
|
|
|
|
| 41 |
"think_token": "<|think|>"
|
| 42 |
},
|
| 43 |
"pad_token": "<pad>",
|
| 44 |
+
"padding_side": "left",
|
| 45 |
"processor_class": "Gemma4Processor",
|
| 46 |
"response_schema": {
|
| 47 |
"properties": {
|
|
|
|
| 82 |
}
|
| 83 |
},
|
| 84 |
"type": "object",
|
| 85 |
+
"x-regex": "(\\<\\|channel\\>thought\\n(?P<thinking>.*?)\\<channel\\|\\>)?(?P<content>(?:(?!\\<\\|tool_call\\>)(?!\\<turn\\|\\>).)+)?(?P<tool_calls>\\<\\|tool_call\\>.*\\<tool_call\\|\\>)?(?:\\<turn\\|\\>)?"
|
| 86 |
},
|
| 87 |
"soc_token": "<|channel>",
|
| 88 |
"sot_token": "<|turn>",
|