Instructions to use google/gemma-4-E4B-it with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use google/gemma-4-E4B-it with Transformers:
# Load model directly from transformers import AutoProcessor, AutoModelForImageTextToText processor = AutoProcessor.from_pretrained("google/gemma-4-E4B-it") model = AutoModelForImageTextToText.from_pretrained("google/gemma-4-E4B-it") - Notebooks
- Google Colab
- Kaggle
ordl-gemma-family
#3
by Winsocket - opened
- README.md +5 -10
- chat_template.jinja +47 -144
- tokenizer_config.json +1 -1
README.md
CHANGED
|
@@ -3,8 +3,6 @@ library_name: transformers
|
|
| 3 |
license: apache-2.0
|
| 4 |
license_link: https://ai.google.dev/gemma/docs/gemma_4_license
|
| 5 |
pipeline_tag: any-to-any
|
| 6 |
-
base_model:
|
| 7 |
-
- google/gemma-4-E4B
|
| 8 |
---
|
| 9 |
|
| 10 |
<div align="center">
|
|
@@ -201,13 +199,13 @@ Once the model is loaded, you can start generating output by directly referencin
|
|
| 201 |
|
| 202 |
|
| 203 |
```python
|
| 204 |
-
# Prompt - add audio
|
| 205 |
messages = [
|
| 206 |
{
|
| 207 |
"role": "user",
|
| 208 |
"content": [
|
|
|
|
| 209 |
{"type": "text", "text": "Transcribe the following speech segment in its original language. Follow these specific instructions for formatting the answer:\n* Only output the transcription, with no newlines.\n* When transcribing numbers, write the digits, i.e. write 1.7 and not one point seven, and write 3 instead of three."},
|
| 210 |
-
{"type": "audio", "audio": "https://raw.githubusercontent.com/google-gemma/cookbook/refs/heads/main/apps/sample-data/journal1.wav"},
|
| 211 |
]
|
| 212 |
}
|
| 213 |
]
|
|
@@ -264,7 +262,7 @@ Once the model is loaded, you can start generating output by directly referencin
|
|
| 264 |
messages = [
|
| 265 |
{
|
| 266 |
"role": "user", "content": [
|
| 267 |
-
{"type": "image", "url": "https://raw.githubusercontent.com/google-gemma/cookbook/refs/heads/main/
|
| 268 |
{"type": "text", "text": "What is shown in this image?"}
|
| 269 |
]
|
| 270 |
}
|
|
@@ -382,10 +380,7 @@ Compared to Gemma 3, the models use standard `system`, `assistant`, and `user` r
|
|
| 382 |
|
| 383 |
### 4. Modality order
|
| 384 |
|
| 385 |
-
For optimal performance with multimodal inputs, place
|
| 386 |
-
|
| 387 |
-
* Image content **before** the text in your prompt.
|
| 388 |
-
* Audio content **after** the text in your prompt.
|
| 389 |
|
| 390 |
### 5. Variable Image Resolution
|
| 391 |
|
|
@@ -517,4 +512,4 @@ The development of vision-language models (VLMs) raises several ethical concerns
|
|
| 517 |
|
| 518 |
### **Benefits**
|
| 519 |
|
| 520 |
-
At the time of release, this family of models provides high-performance open vision-language model implementations designed from the ground up for responsible AI development compared to similarly sized models.
|
|
|
|
| 3 |
license: apache-2.0
|
| 4 |
license_link: https://ai.google.dev/gemma/docs/gemma_4_license
|
| 5 |
pipeline_tag: any-to-any
|
|
|
|
|
|
|
| 6 |
---
|
| 7 |
|
| 8 |
<div align="center">
|
|
|
|
| 199 |
|
| 200 |
|
| 201 |
```python
|
| 202 |
+
# Prompt - add audio before text
|
| 203 |
messages = [
|
| 204 |
{
|
| 205 |
"role": "user",
|
| 206 |
"content": [
|
| 207 |
+
{"type": "audio", "audio": "https://raw.githubusercontent.com/google-gemma/cookbook/refs/heads/main/Demos/sample-data/journal1.wav"},
|
| 208 |
{"type": "text", "text": "Transcribe the following speech segment in its original language. Follow these specific instructions for formatting the answer:\n* Only output the transcription, with no newlines.\n* When transcribing numbers, write the digits, i.e. write 1.7 and not one point seven, and write 3 instead of three."},
|
|
|
|
| 209 |
]
|
| 210 |
}
|
| 211 |
]
|
|
|
|
| 262 |
messages = [
|
| 263 |
{
|
| 264 |
"role": "user", "content": [
|
| 265 |
+
{"type": "image", "url": "https://raw.githubusercontent.com/google-gemma/cookbook/refs/heads/main/Demos/sample-data/GoldenGate.png"},
|
| 266 |
{"type": "text", "text": "What is shown in this image?"}
|
| 267 |
]
|
| 268 |
}
|
|
|
|
| 380 |
|
| 381 |
### 4. Modality order
|
| 382 |
|
| 383 |
+
* For optimal performance with multimodal inputs, place image and/or audio content **before** the text in your prompt.
|
|
|
|
|
|
|
|
|
|
| 384 |
|
| 385 |
### 5. Variable Image Resolution
|
| 386 |
|
|
|
|
| 512 |
|
| 513 |
### **Benefits**
|
| 514 |
|
| 515 |
+
At the time of release, this family of models provides high-performance open vision-language model implementations designed from the ground up for responsible AI development compared to similarly sized models.
|
chat_template.jinja
CHANGED
|
@@ -1,9 +1,9 @@
|
|
| 1 |
-
{%- macro format_parameters(properties, required
|
| 2 |
{%- set standard_keys = ['description', 'type', 'properties', 'required', 'nullable'] -%}
|
| 3 |
{%- set ns = namespace(found_first=false) -%}
|
| 4 |
{%- for key, value in properties | dictsort -%}
|
| 5 |
{%- set add_comma = false -%}
|
| 6 |
-
{%- if
|
| 7 |
{%- if ns.found_first %},{% endif -%}
|
| 8 |
{%- set ns.found_first = true -%}
|
| 9 |
{{ key }}:{
|
|
@@ -11,15 +11,34 @@
|
|
| 11 |
description:<|"|>{{ value['description'] }}<|"|>
|
| 12 |
{%- set add_comma = true -%}
|
| 13 |
{%- endif -%}
|
|
|
|
|
|
|
|
|
|
|
|
|
| 14 |
{%- if value['type'] | upper == 'STRING' -%}
|
| 15 |
{%- if value['enum'] -%}
|
| 16 |
{%- if add_comma %},{%- else -%} {%- set add_comma = true -%} {% endif -%}
|
| 17 |
enum:{{ format_argument(value['enum']) }}
|
| 18 |
{%- endif -%}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 19 |
{%- elif value['type'] | upper == 'ARRAY' -%}
|
| 20 |
{%- if value['items'] is mapping and value['items'] -%}
|
| 21 |
-
|
| 22 |
-
items:{
|
| 23 |
{%- set ns_items = namespace(found_first=false) -%}
|
| 24 |
{%- for item_key, item_value in value['items'] | dictsort -%}
|
| 25 |
{%- if item_value is not none -%}
|
|
@@ -52,32 +71,6 @@
|
|
| 52 |
}
|
| 53 |
{%- endif -%}
|
| 54 |
{%- endif -%}
|
| 55 |
-
{%- if value['nullable'] %}
|
| 56 |
-
{%- if add_comma %},{%- else -%} {%- set add_comma = true -%} {% endif -%}
|
| 57 |
-
nullable:true
|
| 58 |
-
{%- endif -%}
|
| 59 |
-
{%- if value['type'] | upper == 'OBJECT' -%}
|
| 60 |
-
{%- if value['properties'] is defined and value['properties'] is mapping -%}
|
| 61 |
-
{%- if add_comma %},{%- else -%} {%- set add_comma = true -%} {% endif -%}
|
| 62 |
-
properties:{
|
| 63 |
-
{{- format_parameters(value['properties'], value['required'] | default([])) -}}
|
| 64 |
-
}
|
| 65 |
-
{%- elif value is mapping -%}
|
| 66 |
-
{%- if add_comma %},{%- else -%} {%- set add_comma = true -%} {% endif -%}
|
| 67 |
-
properties:{
|
| 68 |
-
{{- format_parameters(value, value['required'] | default([]), filter_keys=true) -}}
|
| 69 |
-
}
|
| 70 |
-
{%- endif -%}
|
| 71 |
-
{%- if value['required'] -%}
|
| 72 |
-
{%- if add_comma %},{%- else -%} {%- set add_comma = true -%} {% endif -%}
|
| 73 |
-
required:[
|
| 74 |
-
{%- for item in value['required'] | default([]) -%}
|
| 75 |
-
<|"|>{{- item -}}<|"|>
|
| 76 |
-
{%- if not loop.last %},{% endif -%}
|
| 77 |
-
{%- endfor -%}
|
| 78 |
-
]
|
| 79 |
-
{%- endif -%}
|
| 80 |
-
{%- endif -%}
|
| 81 |
{%- if add_comma %},{%- else -%} {%- set add_comma = true -%} {% endif -%}
|
| 82 |
type:<|"|>{{ value['type'] | upper }}<|"|>}
|
| 83 |
{%- endif -%}
|
|
@@ -157,42 +150,24 @@
|
|
| 157 |
{{- ns.result | trim -}}
|
| 158 |
{%- endmacro -%}
|
| 159 |
|
| 160 |
-
{%- macro format_tool_response_block(tool_name, response) -%}
|
| 161 |
-
{{- '<|tool_response>' -}}
|
| 162 |
-
{%- if response is mapping -%}
|
| 163 |
-
{{- 'response:' + tool_name + '{' -}}
|
| 164 |
-
{%- for key, value in response | dictsort -%}
|
| 165 |
-
{{- key -}}:{{- format_argument(value, escape_keys=False) -}}
|
| 166 |
-
{%- if not loop.last %},{% endif -%}
|
| 167 |
-
{%- endfor -%}
|
| 168 |
-
{{- '}' -}}
|
| 169 |
-
{%- else -%}
|
| 170 |
-
{{- 'response:' + tool_name + '{value:' + format_argument(response, escape_keys=False) + '}' -}}
|
| 171 |
-
{%- endif -%}
|
| 172 |
-
{{- '<tool_response|>' -}}
|
| 173 |
-
{%- endmacro -%}
|
| 174 |
-
|
| 175 |
{%- set ns = namespace(prev_message_type=None) -%}
|
| 176 |
{%- set loop_messages = messages -%}
|
| 177 |
-
{{
|
| 178 |
{#- Handle System/Tool Definitions Block -#}
|
| 179 |
{%- if (enable_thinking is defined and enable_thinking) or tools or messages[0]['role'] in ['system', 'developer'] -%}
|
| 180 |
{{- '<|turn>system\n' -}}
|
|
|
|
| 181 |
{#- Inject Thinking token at the very top of the FIRST system turn -#}
|
| 182 |
{%- if enable_thinking is defined and enable_thinking -%}
|
| 183 |
-
{{- '<|think|>
|
| 184 |
{%- set ns.prev_message_type = 'think' -%}
|
| 185 |
{%- endif -%}
|
|
|
|
| 186 |
{%- if messages[0]['role'] in ['system', 'developer'] -%}
|
| 187 |
-
{
|
| 188 |
-
{{- messages[0]['content'] | trim -}}
|
| 189 |
-
{%- elif messages[0]['content'] is sequence -%}
|
| 190 |
-
{%- for item in messages[0]['content'] -%}
|
| 191 |
-
{{- item['text'] | trim + ' '-}}
|
| 192 |
-
{%- endfor -%}
|
| 193 |
-
{%- endif -%}
|
| 194 |
{%- set loop_messages = messages[1:] -%}
|
| 195 |
{%- endif -%}
|
|
|
|
| 196 |
{%- if tools -%}
|
| 197 |
{%- for tool in tools %}
|
| 198 |
{{- '<|tool>' -}}
|
|
@@ -201,44 +176,15 @@
|
|
| 201 |
{%- endfor %}
|
| 202 |
{%- set ns.prev_message_type = 'tool' -%}
|
| 203 |
{%- endif -%}
|
|
|
|
| 204 |
{{- '<turn|>\n' -}}
|
| 205 |
{%- endif %}
|
| 206 |
|
| 207 |
-
{#- Pre-scan: find last user message index for reasoning guard -#}
|
| 208 |
-
{%- set ns_turn = namespace(last_user_idx=-1) -%}
|
| 209 |
-
{%- for i in range(loop_messages | length) -%}
|
| 210 |
-
{%- if loop_messages[i]['role'] == 'user' -%}
|
| 211 |
-
{%- set ns_turn.last_user_idx = i -%}
|
| 212 |
-
{%- endif -%}
|
| 213 |
-
{%- endfor -%}
|
| 214 |
-
|
| 215 |
{#- Loop through messages -#}
|
| 216 |
{%- for message in loop_messages -%}
|
| 217 |
-
{%- if message['role'] != 'tool' -%}
|
| 218 |
{%- set ns.prev_message_type = None -%}
|
| 219 |
{%- set role = 'model' if message['role'] == 'assistant' else message['role'] -%}
|
| 220 |
-
{#- Detect continuation: suppress duplicate <|turn>model when previous non-tool message was also assistant -#}
|
| 221 |
-
{%- set prev_nt = namespace(role=None, found=false) -%}
|
| 222 |
-
{%- if loop.index0 > 0 -%}
|
| 223 |
-
{%- for j in range(loop.index0 - 1, -1, -1) -%}
|
| 224 |
-
{%- if not prev_nt.found -%}
|
| 225 |
-
{%- if loop_messages[j]['role'] != 'tool' -%}
|
| 226 |
-
{%- set prev_nt.role = loop_messages[j]['role'] -%}
|
| 227 |
-
{%- set prev_nt.found = true -%}
|
| 228 |
-
{%- endif -%}
|
| 229 |
-
{%- endif -%}
|
| 230 |
-
{%- endfor -%}
|
| 231 |
-
{%- endif -%}
|
| 232 |
-
{%- set continue_same_model_turn = (role == 'model' and prev_nt.role == 'assistant') -%}
|
| 233 |
-
{%- if not continue_same_model_turn -%}
|
| 234 |
{{- '<|turn>' + role + '\n' }}
|
| 235 |
-
{%- endif -%}
|
| 236 |
-
|
| 237 |
-
{#- Render reasoning/reasoning_content as thinking channel -#}
|
| 238 |
-
{%- set thinking_text = message.get('reasoning') or message.get('reasoning_content') -%}
|
| 239 |
-
{%- if thinking_text and loop.index0 > ns_turn.last_user_idx and message.get('tool_calls') -%}
|
| 240 |
-
{{- '<|channel>thought\n' + thinking_text + '\n<channel|>' -}}
|
| 241 |
-
{%- endif -%}
|
| 242 |
|
| 243 |
{%- if message['tool_calls'] -%}
|
| 244 |
{%- for tool_call in message['tool_calls'] -%}
|
|
@@ -259,61 +205,25 @@
|
|
| 259 |
{%- set ns.prev_message_type = 'tool_call' -%}
|
| 260 |
{%- endif -%}
|
| 261 |
|
| 262 |
-
{%-
|
| 263 |
-
|
| 264 |
-
{#- Legacy: tool_responses embedded on the assistant message (Google/Gemma native) -#}
|
| 265 |
{%- for tool_response in message['tool_responses'] -%}
|
| 266 |
-
{{-
|
| 267 |
-
{%-
|
| 268 |
-
|
| 269 |
-
|
| 270 |
-
|
| 271 |
-
|
| 272 |
-
{%- set ns_tool_scan = namespace(stopped=false) -%}
|
| 273 |
-
{%- for k in range(loop.index0 + 1, loop_messages | length) -%}
|
| 274 |
-
{%- if ns_tool_scan.stopped -%}
|
| 275 |
-
{%- elif loop_messages[k]['role'] != 'tool' -%}
|
| 276 |
-
{%- set ns_tool_scan.stopped = true -%}
|
| 277 |
-
{%- else -%}
|
| 278 |
-
{%- set follow = loop_messages[k] -%}
|
| 279 |
-
{#- Resolve tool_call_id to function name -#}
|
| 280 |
-
{%- set ns_tname = namespace(name=follow.get('name') | default('unknown')) -%}
|
| 281 |
-
{%- for tc in message['tool_calls'] -%}
|
| 282 |
-
{%- if tc.get('id') == follow.get('tool_call_id') -%}
|
| 283 |
-
{%- set ns_tname.name = tc['function']['name'] -%}
|
| 284 |
-
{%- endif -%}
|
| 285 |
{%- endfor -%}
|
| 286 |
-
{
|
| 287 |
-
|
| 288 |
-
{
|
| 289 |
-
{{- format_tool_response_block(ns_tname.name, tool_body) -}}
|
| 290 |
-
{%- elif tool_body is sequence and tool_body is not string -%}
|
| 291 |
-
{%- set ns_txt = namespace(s='') -%}
|
| 292 |
-
{%- for part in tool_body -%}
|
| 293 |
-
{%- if part.get('type') == 'text' -%}
|
| 294 |
-
{%- set ns_txt.s = ns_txt.s + (part.get('text') | default('')) -%}
|
| 295 |
-
{%- endif -%}
|
| 296 |
-
{%- endfor -%}
|
| 297 |
-
{{- format_tool_response_block(ns_tname.name, ns_txt.s) -}}
|
| 298 |
-
{%- for part in tool_body -%}
|
| 299 |
-
{%- if part.get('type') == 'image' -%}
|
| 300 |
-
{{- '<|image|>' -}}
|
| 301 |
-
{%- elif part.get('type') == 'audio' -%}
|
| 302 |
-
{{- '<|audio|>' -}}
|
| 303 |
-
{%- elif part.get('type') == 'video' -%}
|
| 304 |
-
{{- '<|video|>' -}}
|
| 305 |
-
{%- endif -%}
|
| 306 |
-
{%- endfor -%}
|
| 307 |
-
{%- else -%}
|
| 308 |
-
{{- format_tool_response_block(ns_tname.name, tool_body) -}}
|
| 309 |
-
{%- endif -%}
|
| 310 |
-
{%- set ns_tr_out.flag = true -%}
|
| 311 |
-
{%- set ns.prev_message_type = 'tool_response' -%}
|
| 312 |
{%- endif -%}
|
|
|
|
| 313 |
{%- endfor -%}
|
|
|
|
| 314 |
{%- endif -%}
|
| 315 |
|
| 316 |
-
{%- set captured_content -%}
|
| 317 |
{%- if message['content'] is string -%}
|
| 318 |
{%- if role == 'model' -%}
|
| 319 |
{{- strip_thinking(message['content']) -}}
|
|
@@ -329,32 +239,25 @@
|
|
| 329 |
{{- item['text'] | trim -}}
|
| 330 |
{%- endif -%}
|
| 331 |
{%- elif item['type'] == 'image' -%}
|
| 332 |
-
{{- '<|image|>' -}}
|
| 333 |
{%- set ns.prev_message_type = 'image' -%}
|
| 334 |
{%- elif item['type'] == 'audio' -%}
|
| 335 |
{{- '<|audio|>' -}}
|
| 336 |
{%- set ns.prev_message_type = 'audio' -%}
|
| 337 |
{%- elif item['type'] == 'video' -%}
|
| 338 |
-
{{- '<|video|>' -}}
|
| 339 |
{%- set ns.prev_message_type = 'video' -%}
|
| 340 |
{%- endif -%}
|
| 341 |
{%- endfor -%}
|
| 342 |
{%- endif -%}
|
| 343 |
-
{%- endset -%}
|
| 344 |
|
| 345 |
-
|
| 346 |
-
{%- set has_content = captured_content | trim | length > 0 -%}
|
| 347 |
-
|
| 348 |
-
{%- if ns.prev_message_type == 'tool_call' and not ns_tr_out.flag -%}
|
| 349 |
-
{{- '<|tool_response>' -}}
|
| 350 |
-
{%- elif not (ns_tr_out.flag and not has_content) -%}
|
| 351 |
{{- '<turn|>\n' -}}
|
| 352 |
{%- endif -%}
|
| 353 |
-
{%- endif -%}
|
| 354 |
{%- endfor -%}
|
| 355 |
|
| 356 |
{%- if add_generation_prompt -%}
|
| 357 |
-
{%- if ns.prev_message_type != 'tool_response'
|
| 358 |
{{- '<|turn>model\n' -}}
|
| 359 |
{%- endif -%}
|
| 360 |
{%- endif -%}
|
|
|
|
| 1 |
+
{%- macro format_parameters(properties, required) -%}
|
| 2 |
{%- set standard_keys = ['description', 'type', 'properties', 'required', 'nullable'] -%}
|
| 3 |
{%- set ns = namespace(found_first=false) -%}
|
| 4 |
{%- for key, value in properties | dictsort -%}
|
| 5 |
{%- set add_comma = false -%}
|
| 6 |
+
{%- if key not in standard_keys -%}
|
| 7 |
{%- if ns.found_first %},{% endif -%}
|
| 8 |
{%- set ns.found_first = true -%}
|
| 9 |
{{ key }}:{
|
|
|
|
| 11 |
description:<|"|>{{ value['description'] }}<|"|>
|
| 12 |
{%- set add_comma = true -%}
|
| 13 |
{%- endif -%}
|
| 14 |
+
{%- if value['nullable'] %}
|
| 15 |
+
{%- if add_comma %},{%- else -%} {%- set add_comma = true -%} {% endif -%}
|
| 16 |
+
nullable:true
|
| 17 |
+
{%- endif -%}
|
| 18 |
{%- if value['type'] | upper == 'STRING' -%}
|
| 19 |
{%- if value['enum'] -%}
|
| 20 |
{%- if add_comma %},{%- else -%} {%- set add_comma = true -%} {% endif -%}
|
| 21 |
enum:{{ format_argument(value['enum']) }}
|
| 22 |
{%- endif -%}
|
| 23 |
+
{%- elif value['type'] | upper == 'OBJECT' -%}
|
| 24 |
+
,properties:{
|
| 25 |
+
{%- if value['properties'] is defined and value['properties'] is mapping -%}
|
| 26 |
+
{{- format_parameters(value['properties'], value['required'] | default([])) -}}
|
| 27 |
+
{%- elif value is mapping -%}
|
| 28 |
+
{{- format_parameters(value, value['required'] | default([])) -}}
|
| 29 |
+
{%- endif -%}
|
| 30 |
+
}
|
| 31 |
+
{%- if value['required'] -%}
|
| 32 |
+
,required:[
|
| 33 |
+
{%- for item in value['required'] | default([]) -%}
|
| 34 |
+
<|"|>{{- item -}}<|"|>
|
| 35 |
+
{%- if not loop.last %},{% endif -%}
|
| 36 |
+
{%- endfor -%}
|
| 37 |
+
]
|
| 38 |
+
{%- endif -%}
|
| 39 |
{%- elif value['type'] | upper == 'ARRAY' -%}
|
| 40 |
{%- if value['items'] is mapping and value['items'] -%}
|
| 41 |
+
,items:{
|
|
|
|
| 42 |
{%- set ns_items = namespace(found_first=false) -%}
|
| 43 |
{%- for item_key, item_value in value['items'] | dictsort -%}
|
| 44 |
{%- if item_value is not none -%}
|
|
|
|
| 71 |
}
|
| 72 |
{%- endif -%}
|
| 73 |
{%- endif -%}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 74 |
{%- if add_comma %},{%- else -%} {%- set add_comma = true -%} {% endif -%}
|
| 75 |
type:<|"|>{{ value['type'] | upper }}<|"|>}
|
| 76 |
{%- endif -%}
|
|
|
|
| 150 |
{{- ns.result | trim -}}
|
| 151 |
{%- endmacro -%}
|
| 152 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 153 |
{%- set ns = namespace(prev_message_type=None) -%}
|
| 154 |
{%- set loop_messages = messages -%}
|
| 155 |
+
{{ bos_token }}
|
| 156 |
{#- Handle System/Tool Definitions Block -#}
|
| 157 |
{%- if (enable_thinking is defined and enable_thinking) or tools or messages[0]['role'] in ['system', 'developer'] -%}
|
| 158 |
{{- '<|turn>system\n' -}}
|
| 159 |
+
|
| 160 |
{#- Inject Thinking token at the very top of the FIRST system turn -#}
|
| 161 |
{%- if enable_thinking is defined and enable_thinking -%}
|
| 162 |
+
{{- '<|think|>' -}}
|
| 163 |
{%- set ns.prev_message_type = 'think' -%}
|
| 164 |
{%- endif -%}
|
| 165 |
+
|
| 166 |
{%- if messages[0]['role'] in ['system', 'developer'] -%}
|
| 167 |
+
{{- messages[0]['content'] | trim -}}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 168 |
{%- set loop_messages = messages[1:] -%}
|
| 169 |
{%- endif -%}
|
| 170 |
+
|
| 171 |
{%- if tools -%}
|
| 172 |
{%- for tool in tools %}
|
| 173 |
{{- '<|tool>' -}}
|
|
|
|
| 176 |
{%- endfor %}
|
| 177 |
{%- set ns.prev_message_type = 'tool' -%}
|
| 178 |
{%- endif -%}
|
| 179 |
+
|
| 180 |
{{- '<turn|>\n' -}}
|
| 181 |
{%- endif %}
|
| 182 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 183 |
{#- Loop through messages -#}
|
| 184 |
{%- for message in loop_messages -%}
|
|
|
|
| 185 |
{%- set ns.prev_message_type = None -%}
|
| 186 |
{%- set role = 'model' if message['role'] == 'assistant' else message['role'] -%}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 187 |
{{- '<|turn>' + role + '\n' }}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 188 |
|
| 189 |
{%- if message['tool_calls'] -%}
|
| 190 |
{%- for tool_call in message['tool_calls'] -%}
|
|
|
|
| 205 |
{%- set ns.prev_message_type = 'tool_call' -%}
|
| 206 |
{%- endif -%}
|
| 207 |
|
| 208 |
+
{%- if message['tool_responses'] -%}
|
| 209 |
+
{#- Tool Response handling -#}
|
|
|
|
| 210 |
{%- for tool_response in message['tool_responses'] -%}
|
| 211 |
+
{{- '<|tool_response>' -}}
|
| 212 |
+
{%- if tool_response['response'] is mapping -%}
|
| 213 |
+
{{- 'response:' + tool_response['name'] | default('unknown') + '{' -}}
|
| 214 |
+
{%- for key, value in tool_response['response'] | dictsort -%}
|
| 215 |
+
{{- key -}}:{{- format_argument(value, escape_keys=False) -}}
|
| 216 |
+
{%- if not loop.last %},{% endif -%}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 217 |
{%- endfor -%}
|
| 218 |
+
{{- '}' -}}
|
| 219 |
+
{%- else -%}
|
| 220 |
+
{{- 'response:' + tool_response['name'] | default('unknown') + '{value:' + format_argument(tool_response['response'], escape_keys=False) + '}' -}}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 221 |
{%- endif -%}
|
| 222 |
+
{{- '<tool_response|>' -}}
|
| 223 |
{%- endfor -%}
|
| 224 |
+
{%- set ns.prev_message_type = 'tool_response' -%}
|
| 225 |
{%- endif -%}
|
| 226 |
|
|
|
|
| 227 |
{%- if message['content'] is string -%}
|
| 228 |
{%- if role == 'model' -%}
|
| 229 |
{{- strip_thinking(message['content']) -}}
|
|
|
|
| 239 |
{{- item['text'] | trim -}}
|
| 240 |
{%- endif -%}
|
| 241 |
{%- elif item['type'] == 'image' -%}
|
| 242 |
+
{{- '\n\n<|image|>\n\n' -}}
|
| 243 |
{%- set ns.prev_message_type = 'image' -%}
|
| 244 |
{%- elif item['type'] == 'audio' -%}
|
| 245 |
{{- '<|audio|>' -}}
|
| 246 |
{%- set ns.prev_message_type = 'audio' -%}
|
| 247 |
{%- elif item['type'] == 'video' -%}
|
| 248 |
+
{{- '\n\n<|video|>\n\n' -}}
|
| 249 |
{%- set ns.prev_message_type = 'video' -%}
|
| 250 |
{%- endif -%}
|
| 251 |
{%- endfor -%}
|
| 252 |
{%- endif -%}
|
|
|
|
| 253 |
|
| 254 |
+
{%- if not (message['tool_responses'] and not message['content']) -%}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 255 |
{{- '<turn|>\n' -}}
|
| 256 |
{%- endif -%}
|
|
|
|
| 257 |
{%- endfor -%}
|
| 258 |
|
| 259 |
{%- if add_generation_prompt -%}
|
| 260 |
+
{%- if ns.prev_message_type != 'tool_response' -%}
|
| 261 |
{{- '<|turn>model\n' -}}
|
| 262 |
{%- endif -%}
|
| 263 |
{%- endif -%}
|
tokenizer_config.json
CHANGED
|
@@ -61,7 +61,7 @@
|
|
| 61 |
}
|
| 62 |
}
|
| 63 |
},
|
| 64 |
-
"x-regex": "(\\<\\|channel\\>thought\\n(?P<thinking>.*?)\\<channel\\|\\>)?(?P<
|
| 65 |
},
|
| 66 |
"soc_token": "<|channel>",
|
| 67 |
"sot_token": "<|turn>",
|
|
|
|
| 61 |
}
|
| 62 |
}
|
| 63 |
},
|
| 64 |
+
"x-regex": "(\\<\\|channel\\>thought\\n(?P<thinking>.*?)\\<channel\\|\\>)?(?P<content>(?:(?!\\<\\|tool_call\\>)(?!\\<turn\\|\\>).)+)?(?P<tool_calls>\\<\\|tool_call\\>.*\\<tool_call\\|\\>)?(?:\\<turn\\|\\>)?"
|
| 65 |
},
|
| 66 |
"soc_token": "<|channel>",
|
| 67 |
"sot_token": "<|turn>",
|