xinference 1.2.0__py3-none-any.whl → 1.2.2__py3-none-any.whl

xinference/model/llm/llm_family_modelscope.json

@@ -4769,10 +4769,11 @@
         "model_format":"mlx",
         "model_size_in_billions":2,
         "quantizations":[
+          "4bit",
           "8bit"
         ],
         "model_hub": "modelscope",
-        "model_id":"okwinds/Qwen2-VL-2B-Instruct-MLX-8bit",
+        "model_id":"mlx-community/Qwen2-VL-2B-Instruct-{quantization}",
         "model_revision":"master"
       },
       {
@@ -4825,6 +4826,97 @@
       "<|endoftext|>"
     ]
   },
+  {
+    "version":1,
+    "context_length":128000,
+    "model_name":"qwen2.5-vl-instruct",
+    "model_lang":[
+      "en",
+      "zh"
+    ],
+    "model_ability":[
+      "chat",
+      "vision"
+    ],
+    "model_description":"Qwen2.5-VL: Qwen2.5-VL is the latest version of the vision language models in the Qwen model familities.",
+    "model_specs":[
+      {
+        "model_format":"pytorch",
+        "model_size_in_billions":3,
+        "quantizations":[
+          "none"
+        ],
+        "model_hub": "modelscope",
+        "model_id":"qwen/Qwen2.5-VL-3B-Instruct"
+      },
+      {
+        "model_format":"pytorch",
+        "model_size_in_billions":7,
+        "quantizations":[
+          "none"
+        ],
+        "model_hub": "modelscope",
+        "model_id":"qwen/Qwen2.5-VL-7B-Instruct"
+      },
+      {
+        "model_format":"pytorch",
+        "model_size_in_billions":72,
+        "quantizations":[
+          "none"
+        ],
+        "model_hub": "modelscope",
+        "model_id":"qwen/Qwen2.5-VL-72B-Instruct"
+      },
+      {
+        "model_format":"mlx",
+        "model_size_in_billions":3,
+        "quantizations":[
+          "3bit",
+          "4bit",
+          "6bit",
+          "8bit",
+          "bf16"
+        ],
+        "model_hub": "modelscope",
+        "model_id":"mlx-community/Qwen2.5-VL-3B-Instruct-{quantization}"
+      },
+      {
+        "model_format":"mlx",
+        "model_size_in_billions":7,
+        "quantizations":[
+          "3bit",
+          "4bit",
+          "6bit",
+          "8bit",
+          "bf16"
+        ],
+        "model_hub": "modelscope",
+        "model_id":"mlx-community/Qwen2.5-VL-7B-Instruct-{quantization}"
+      },
+      {
+        "model_format":"mlx",
+        "model_size_in_billions":72,
+        "quantizations":[
+          "3bit",
+          "4bit",
+          "6bit",
+          "8bit",
+          "bf16"
+        ],
+        "model_hub": "modelscope",
+        "model_id":"mlx-community/Qwen2.5-VL-72B-Instruct-{quantization}"
+      }
+    ],
+    "chat_template": "{% set image_count = namespace(value=0) %}{% set video_count = namespace(value=0) %}{% for message in messages %}{% if loop.first and message['role'] != 'system' %}<|im_start|>system\nYou are a helpful assistant.<|im_end|>\n{% endif %}<|im_start|>{{ message['role'] }}\n{% if message['content'] is string %}{{ message['content'] }}<|im_end|>\n{% else %}{% for content in message['content'] %}{% if content['type'] == 'image' or 'image' in content or 'image_url' in content %}{% set image_count.value = image_count.value + 1 %}{% if add_vision_id %}Picture {{ image_count.value }}: {% endif %}<|vision_start|><|image_pad|><|vision_end|>{% elif content['type'] == 'video' or 'video' in content %}{% set video_count.value = video_count.value + 1 %}{% if add_vision_id %}Video {{ video_count.value }}: {% endif %}<|vision_start|><|video_pad|><|vision_end|>{% elif 'text' in content %}{{ content['text'] }}{% endif %}{% endfor %}<|im_end|>\n{% endif %}{% endfor %}{% if add_generation_prompt %}<|im_start|>assistant\n{% endif %}",
+    "stop_token_ids": [
+      151645,
+      151643
+    ],
+    "stop": [
+      "<|im_end|>",
+      "<|endoftext|>"
+    ]
+  },
   {
     "version": 1,
     "context_length": 32768,
@@ -5558,7 +5650,7 @@
           "q8_0"
         ],
         "model_id": "qwen/Qwen2.5-7B-Instruct-GGUF",
-        "model_file_name_template": "qwen2_5-7b-instruct-{quantization}.gguf",
+        "model_file_name_template": "qwen2.5-7b-instruct-{quantization}.gguf",
         "model_hub": "modelscope",
         "model_file_name_split_template": "qwen2.5-7b-instruct-{quantization}-{part}.gguf",
         "quantization_parts": {
@@ -6433,6 +6525,326 @@
       "<|im_end|>"
     ]
   },
+  {
+    "version": 1,
+    "context_length": 131072,
+    "model_name": "deepseek-r1-distill-qwen",
+    "model_lang": [
+      "en",
+      "zh"
+    ],
+    "model_ability": [
+      "chat"
+    ],
+    "model_description": "deepseek-r1-distill-qwen is distilled from DeepSeek-R1 based on Qwen",
+    "model_specs": [
+      {
+        "model_format": "pytorch",
+        "model_size_in_billions": "1_5",
+        "quantizations": [
+          "4-bit",
+          "8-bit",
+          "none"
+        ],
+        "model_id": "deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B",
+        "model_hub": "modelscope"
+      },
+      {
+        "model_format": "ggufv2",
+        "model_size_in_billions": "1_5",
+        "quantizations": [
+          "Q2_K",
+          "Q2_K_L",
+          "Q3_K_M",
+          "Q4_K_M",
+          "Q5_K_M",
+          "Q6_K",
+          "Q8_0"
+        ],
+        "model_id": "unsloth/DeepSeek-R1-Distill-Qwen-1.5B-GGUF",
+        "model_file_name_template": "DeepSeek-R1-Distill-Qwen-1.5B-{quantization}.gguf",
+        "model_hub": "modelscope"
+      },
+      {
+        "model_format": "mlx",
+        "model_size_in_billions": "1_5",
+        "quantizations": [
+          "3bit",
+          "4bit",
+          "6bit",
+          "8bit",
+          "bf16"
+        ],
+        "model_id": "mlx-community/DeepSeek-R1-Distill-Qwen-1.5B-{quantization}",
+        "model_hub": "modelscope"
+      },
+      {
+        "model_format": "pytorch",
+        "model_size_in_billions": 7,
+        "quantizations": [
+          "4-bit",
+          "8-bit",
+          "none"
+        ],
+        "model_id": "deepseek-ai/DeepSeek-R1-Distill-Qwen-7B",
+        "model_hub": "modelscope"
+      },
+      {
+        "model_format": "gptq",
+        "model_size_in_billions": 7,
+        "quantizations": [
+          "Int4"
+        ],
+        "model_id": "tclf90/deepseek-r1-distill-qwen-7b-gptq-int4",
+        "model_hub": "modelscope"
+      },
+      {
+        "model_format": "ggufv2",
+        "model_size_in_billions": 7,
+        "quantizations": [
+          "Q2_K",
+          "Q2_K_L",
+          "Q3_K_M",
+          "Q4_K_M",
+          "Q5_K_M",
+          "Q6_K",
+          "Q8_0",
+          "F16"
+        ],
+        "model_id": "unsloth/DeepSeek-R1-Distill-Qwen-7B-GGUF",
+        "model_file_name_template": "DeepSeek-R1-Distill-Qwen-7B-{quantization}.gguf",
+        "model_hub": "modelscope"
+      },
+      {
+        "model_format": "mlx",
+        "model_size_in_billions": 7,
+        "quantizations": [
+          "3bit",
+          "4bit",
+          "6bit",
+          "8bit"
+        ],
+        "model_id": "okwinds/DeepSeek-R1-Distill-Qwen-7B-MLX-{quantization}",
+        "model_hub": "modelscope"
+      },
+      {
+        "model_format": "pytorch",
+        "model_size_in_billions": 14,
+        "quantizations": [
+          "4-bit",
+          "8-bit",
+          "none"
+        ],
+        "model_id": "deepseek-ai/DeepSeek-R1-Distill-Qwen-14B",
+        "model_hub": "modelscope"
+      },
+      {
+        "model_format": "ggufv2",
+        "model_size_in_billions": 14,
+        "quantizations": [
+          "Q2_K",
+          "Q2_K_L",
+          "Q3_K_M",
+          "Q4_K_M",
+          "Q5_K_M",
+          "Q6_K",
+          "Q8_0",
+          "F16"
+        ],
+        "model_id": "unsloth/DeepSeek-R1-Distill-Qwen-14B-GGUF",
+        "model_file_name_template": "DeepSeek-R1-Distill-Qwen-14B-{quantization}.gguf",
+        "model_hub": "modelscope"
+      },
+      {
+        "model_format": "mlx",
+        "model_size_in_billions": 14,
+        "quantizations": [
+          "3bit",
+          "4bit",
+          "6bit",
+          "8bit"
+        ],
+        "model_id": "okwinds/DeepSeek-R1-Distill-Qwen-14B-MLX-{quantization}",
+        "model_hub": "modelscope"
+      },
+      {
+        "model_format": "pytorch",
+        "model_size_in_billions": 32,
+        "quantizations": [
+          "4-bit",
+          "8-bit",
+          "none"
+        ],
+        "model_id": "deepseek-ai/DeepSeek-R1-Distill-Qwen-32B",
+        "model_hub": "modelscope"
+      },
+      {
+        "model_format": "gptq",
+        "model_size_in_billions": 32,
+        "quantizations": [
+          "Int4"
+        ],
+        "model_id": "tclf90/deepseek-r1-distill-qwen-32b-gptq-int4",
+        "model_hub": "modelscope"
+      },
+      {
+        "model_format": "ggufv2",
+        "model_size_in_billions": 32,
+        "quantizations": [
+          "Q2_K",
+          "Q2_K_L",
+          "Q3_K_M",
+          "Q4_K_M",
+          "Q5_K_M",
+          "Q6_K",
+          "Q8_0",
+          "F16"
+        ],
+        "model_id": "unsloth/DeepSeek-R1-Distill-Qwen-32B-GGUF",
+        "model_file_name_template": "DeepSeek-R1-Distill-Qwen-32B-{quantization}.gguf",
+        "model_hub": "modelscope"
+      },
+      {
+        "model_format": "mlx",
+        "model_size_in_billions": 32,
+        "quantizations": [
+          "2bit",
+          "3bit",
+          "4bit",
+          "6bit",
+          "8bit"
+        ],
+        "model_id": "okwinds/DeepSeek-R1-Distill-Qwen-32B-MLX-{quantization}",
+        "model_hub": "modelscope"
+      }
+    ],
+    "chat_template": "{% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% set ns = namespace(is_first=false, is_tool=false, is_output_first=true, system_prompt='') %}{%- for message in messages %}{%- if message['role'] == 'system' %}{% set ns.system_prompt = message['content'] %}{%- endif %}{%- endfor %}{{bos_token}}{{ns.system_prompt}}{%- for message in messages %}{%- if message['role'] == 'user' %}{%- set ns.is_tool = false -%}{{'<｜User｜>' + message['content']}}{%- endif %}{%- if message['role'] == 'assistant' and message['content'] is none %}{%- set ns.is_tool = false -%}{%- for tool in message['tool_calls']%}{%- if not ns.is_first %}{{'<｜Assistant｜><｜tool▁calls▁begin｜><｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '\\n' + '```json' + '\\n' + tool['function']['arguments'] + '\\n' + '```' + '<｜tool▁call▁end｜>'}}{%- set ns.is_first = true -%}{%- else %}{{'\\n' + '<｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '\\n' + '```json' + '\\n' + tool['function']['arguments'] + '\\n' + '```' + '<｜tool▁call▁end｜>'}}{{'<｜tool▁calls▁end｜><｜end▁of▁sentence｜>'}}{%- endif %}{%- endfor %}{%- endif %}{%- if message['role'] == 'assistant' and message['content'] is not none %}{%- if ns.is_tool %}{{'<｜tool▁outputs▁end｜>' + message['content'] + '<｜end▁of▁sentence｜>'}}{%- set ns.is_tool = false -%}{%- else %}{% set content = message['content'] %}{% if '</think>' in content %}{% set content = content.split('</think>')[-1] %}{% endif %}{{'<｜Assistant｜>' + content + '<｜end▁of▁sentence｜>'}}{%- endif %}{%- endif %}{%- if message['role'] == 'tool' %}{%- set ns.is_tool = true -%}{%- if ns.is_output_first %}{{'<｜tool▁outputs▁begin｜><｜tool▁output▁begin｜>' + message['content'] + '<｜tool▁output▁end｜>'}}{%- set ns.is_output_first = false %}{%- else %}{{'\\n<｜tool▁output▁begin｜>' + message['content'] + '<｜tool▁output▁end｜>'}}{%- endif %}{%- endif %}{%- endfor -%}{% if ns.is_tool %}{{'<｜tool▁outputs▁end｜>'}}{% endif %}{% if add_generation_prompt and not ns.is_tool %}{{'<｜Assistant｜>'}}{% endif %}",
+    "stop_token_ids": [
+      151643
+    ],
+    "stop": [
+      "<｜end▁of▁sentence｜>"
+    ]
+  },
+  {
+    "version": 1,
+    "context_length": 131072,
+    "model_name": "deepseek-r1-distill-llama",
+    "model_lang": [
+      "en",
+      "zh"
+    ],
+    "model_ability": [
+      "chat"
+    ],
+    "model_description": "deepseek-r1-distill-llama is distilled from DeepSeek-R1 based on Llama",
+    "model_specs": [
+      {
+        "model_format": "pytorch",
+        "model_size_in_billions": 8,
+        "quantizations": [
+          "4-bit",
+          "8-bit",
+          "none"
+        ],
+        "model_id": "deepseek-ai/DeepSeek-R1-Distill-Llama-8B",
+        "model_hub": "modelscope"
+      },
+      {
+        "model_format": "ggufv2",
+        "model_size_in_billions": 8,
+        "quantizations": [
+          "Q2_K",
+          "Q2_K_L",
+          "Q3_K_M",
+          "Q4_K_M",
+          "Q5_K_M",
+          "Q6_K",
+          "Q8_0",
+          "F16"
+        ],
+        "model_id": "unsloth/DeepSeek-R1-Distill-Llama-8B-GGUF",
+        "model_file_name_template": "DeepSeek-R1-Distill-Llama-8B-{quantization}.gguf",
+        "model_hub": "modelscope"
+      },
+      {
+        "model_format": "mlx",
+        "model_size_in_billions": 8,
+        "quantizations": [
+          "3bit",
+          "4bit",
+          "6bit",
+          "8bit",
+          "bf16"
+        ],
+        "model_id": "okwinds/DeepSeek-R1-Distill-Llama-8B-MLX-{quantization}",
+        "model_hub": "modelscope"
+      },
+      {
+        "model_format": "pytorch",
+        "model_size_in_billions": 70,
+        "quantizations": [
+          "4-bit",
+          "8-bit",
+          "none"
+        ],
+        "model_id": "deepseek-ai/DeepSeek-R1-Distill-Llama-70B",
+        "model_hub": "modelscope"
+      },
+      {
+        "model_format": "ggufv2",
+        "model_size_in_billions": 70,
+        "quantizations": [
+          "Q2_K",
+          "Q2_K_L",
+          "Q3_K_M",
+          "Q4_K_M",
+          "Q5_K_M",
+          "Q6_K",
+          "Q8_0",
+          "F16"
+        ],
+        "quantization_parts": {
+          "Q6_K": [
+            "00001-of-00002",
+            "00002-of-00002"
+          ],
+          "Q8_0": [
+            "00001-of-00002",
+            "00002-of-00002"
+          ],
+          "F16": [
+            "00001-of-00003",
+            "00002-of-00003",
+            "00003-of-00003"
+          ]
+        },
+        "model_id": "unsloth/DeepSeek-R1-Distill-Llama-70B-GGUF",
+        "model_file_name_template": "DeepSeek-R1-Distill-Qwen-7B-{quantization}.gguf",
+        "model_file_name_split_template": "DeepSeek-R1-Distill-Llama-70B-{quantization}/DeepSeek-R1-Distill-Llama-70B-{quantization}-{part}.gguf",
+        "model_hub": "modelscope"
+      },
+      {
+        "model_format": "mlx",
+        "model_size_in_billions": 70,
+        "quantizations": [
+          "3bit",
+          "4bit",
+          "6bit",
+          "8bit"
+        ],
+        "model_id": "okwinds/DeepSeek-R1-Distill-Llama-70B-MLX-{quantization}",
+        "model_hub": "modelscope"
+      }
+    ],
+    "chat_template": "{% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% set ns = namespace(is_first=false, is_tool=false, is_output_first=true, system_prompt='') %}{%- for message in messages %}{%- if message['role'] == 'system' %}{% set ns.system_prompt = message['content'] %}{%- endif %}{%- endfor %}{{bos_token}}{{ns.system_prompt}}{%- for message in messages %}{%- if message['role'] == 'user' %}{%- set ns.is_tool = false -%}{{'<｜User｜>' + message['content']}}{%- endif %}{%- if message['role'] == 'assistant' and message['content'] is none %}{%- set ns.is_tool = false -%}{%- for tool in message['tool_calls']%}{%- if not ns.is_first %}{{'<｜Assistant｜><｜tool▁calls▁begin｜><｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '\\n' + '```json' + '\\n' + tool['function']['arguments'] + '\\n' + '```' + '<｜tool▁call▁end｜>'}}{%- set ns.is_first = true -%}{%- else %}{{'\\n' + '<｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '\\n' + '```json' + '\\n' + tool['function']['arguments'] + '\\n' + '```' + '<｜tool▁call▁end｜>'}}{{'<｜tool▁calls▁end｜><｜end▁of▁sentence｜>'}}{%- endif %}{%- endfor %}{%- endif %}{%- if message['role'] == 'assistant' and message['content'] is not none %}{%- if ns.is_tool %}{{'<｜tool▁outputs▁end｜>' + message['content'] + '<｜end▁of▁sentence｜>'}}{%- set ns.is_tool = false -%}{%- else %}{% set content = message['content'] %}{% if '</think>' in content %}{% set content = content.split('</think>')[-1] %}{% endif %}{{'<｜Assistant｜>' + content + '<｜end▁of▁sentence｜>'}}{%- endif %}{%- endif %}{%- if message['role'] == 'tool' %}{%- set ns.is_tool = true -%}{%- if ns.is_output_first %}{{'<｜tool▁outputs▁begin｜><｜tool▁output▁begin｜>' + message['content'] + '<｜tool▁output▁end｜>'}}{%- set ns.is_output_first = false %}{%- else %}{{'\\n<｜tool▁output▁begin｜>' + message['content'] + '<｜tool▁output▁end｜>'}}{%- endif %}{%- endif %}{%- endfor -%}{% if ns.is_tool %}{{'<｜tool▁outputs▁end｜>'}}{% endif %}{% if add_generation_prompt and not ns.is_tool %}{{'<｜Assistant｜>'}}{% endif %}",
+    "stop_token_ids": [
+      151643
+    ],
+    "stop": [
+      "<｜end▁of▁sentence｜>"
+    ]
+  },
   {
     "version": 1,
     "context_length": 8192,
@@ -6723,7 +7135,7 @@
       "<|endoftext|>"
     ]
   },
-   {
+  {
     "version": 1,
     "context_length": 32768,
     "model_name": "marco-o1",
@@ -6821,5 +7233,85 @@
       "<|user|>",
       "<|observation|>"
     ]
+  },
+  {
+    "version": 1,
+    "context_length": 32768,
+    "model_name": "internlm3-instruct",
+    "model_lang": [
+      "en",
+      "zh"
+    ],
+    "model_ability": [
+      "chat",
+      "tools"
+    ],
+    "model_description": "InternLM3 has open-sourced an 8-billion parameter instruction model, InternLM3-8B-Instruct, designed for general-purpose usage and advanced reasoning.",
+    "model_specs": [
+      {
+        "model_format": "pytorch",
+        "model_size_in_billions": 8,
+        "quantizations": [
+          "4-bit",
+          "8-bit",
+          "none"
+        ],
+        "model_id": "Shanghai_AI_Laboratory/internlm3-8b-instruct",
+        "model_hub": "modelscope"
+      },
+      {
+        "model_format": "gptq",
+        "model_size_in_billions": 8,
+        "quantizations": [
+          "Int4"
+        ],
+        "model_id": "Shanghai_AI_Laboratory/internlm3-8b-instruct-gptq-int4",
+        "model_hub": "modelscope"
+      },
+      {
+        "model_format": "awq",
+        "model_size_in_billions": 8,
+        "quantizations": [
+          "Int4"
+        ],
+        "model_id": "Shanghai_AI_Laboratory/internlm3-8b-instruct-awq",
+        "model_hub": "modelscope"
+      },
+      {
+        "model_format": "ggufv2",
+        "model_size_in_billions": 8,
+        "quantizations": [
+          "q2_k",
+          "q3_k_m",
+          "q4_0",
+          "q4_k_m",
+          "q5_0",
+          "q5_k_m",
+          "q6_k",
+          "q8_0"
+        ],
+        "model_id": "Shanghai_AI_Laboratory/internlm3-8b-instruct-gguf",
+        "model_file_name_template": "internlm3-8b-instruct-{quantization}.gguf",
+        "model_hub": "modelscope"
+      },
+      {
+        "model_format":"mlx",
+        "model_size_in_billions":8,
+        "quantizations":[
+          "4bit"
+        ],
+        "model_hub": "modelscope",
+        "model_id":"mlx-community/internlm3-8b-instruct-{quantization}"
+      }
+    ],
+    "chat_template": "{{ bos_token }}{% for message in messages %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}",
+    "stop_token_ids": [
+      2,
+      128131
+    ],
+    "stop": [
+      "</s>",
+      "<|im_end|>"
+    ]
   }
 ]

xinference/model/llm/memory.py

@@ -14,7 +14,7 @@
 
 # NOTE:
 #
-#   The algorithum is ported from https://github.com/RahulSChand/gpu_poor
+#   The algorithm is ported from https://github.com/RahulSChand/gpu_poor
 #
 #   Improvement:
 #

xinference/model/llm/mlx/core.py

@@ -31,7 +31,12 @@ from ....types import (
 )
 from ..core import LLM
 from ..llm_family import LLMFamilyV1, LLMSpecV1
-from ..utils import QWEN_TOOL_CALL_FAMILY, ChatModelMixin, generate_completion_chunk
+from ..utils import (
+    DEEPSEEK_TOOL_CALL_FAMILY,
+    QWEN_TOOL_CALL_FAMILY,
+    ChatModelMixin,
+    generate_completion_chunk,
+)
 
 logger = logging.getLogger(__name__)
 
@@ -103,10 +108,10 @@ class MLXModel(LLM):
         # default config is adapted from
         # https://github.com/ml-explore/mlx-examples/blob/f212b770d8b5143e23102eda20400ae43340f844/llms/mlx_lm/utils.py#L129
         generate_config.setdefault("temperature", 0.0)
+        generate_config.setdefault("logit_bias", None)
         generate_config.setdefault("repetition_penalty", None)
         generate_config.setdefault("repetition_context_size", 20)
         generate_config.setdefault("top_p", 1.0)
-        generate_config.setdefault("logit_bias", None)
         return generate_config
 
     def _load_model(self, **kwargs):
@@ -199,14 +204,24 @@ class MLXModel(LLM):
         return prompt
 
     def _generate_stream_inner(self, **kwargs):
-        from mlx_lm.utils import make_sampler, stream_generate
+        from mlx_lm.utils import make_logits_processors, make_sampler, stream_generate
 
         sampler = make_sampler(
             temp=kwargs.pop("temperature"), top_p=kwargs.pop("top_p")
         )
         prompt_token_ids = kwargs.pop("prompt_token_ids")
+        logits_processors = make_logits_processors(
+            logit_bias=kwargs.pop("logits_bias", None),
+            repetition_penalty=kwargs.pop("repetition_penalty"),
+            repetition_context_size=kwargs.pop("repetition_context_size"),
+        )
         yield from stream_generate(
-            self._model, self._tokenizer, prompt_token_ids, sampler=sampler, **kwargs
+            self._model,
+            self._tokenizer,
+            prompt_token_ids,
+            sampler=sampler,
+            logits_processors=logits_processors,
+            **kwargs,
         )
 
     def _prepare_inputs(
@@ -414,8 +429,11 @@ class MLXChatModel(MLXModel, ChatModelMixin):
         model_family = self.model_family.model_family or self.model_family.model_name
         tools = generate_config.pop("tools", []) if generate_config else None
         full_context_kwargs = {}
-        if tools and model_family in QWEN_TOOL_CALL_FAMILY:
-            full_context_kwargs["tools"] = tools
+        if tools:
+            if model_family in QWEN_TOOL_CALL_FAMILY:
+                full_context_kwargs["tools"] = tools
+            elif model_family in DEEPSEEK_TOOL_CALL_FAMILY:
+                self._tools_to_messages_for_deepseek(messages, tools)
         assert self.model_family.chat_template is not None
         full_prompt = self.get_full_context(
             messages, self.model_family.chat_template, **full_context_kwargs

xinference/model/llm/transformers/core.py

@@ -39,7 +39,12 @@ from ....types import (
 from ...utils import select_device
 from ..core import LLM
 from ..llm_family import LLMFamilyV1, LLMSpecV1
-from ..utils import LLAMA3_TOOL_CALL_FAMILY, QWEN_TOOL_CALL_FAMILY, ChatModelMixin
+from ..utils import (
+    DEEPSEEK_TOOL_CALL_FAMILY,
+    LLAMA3_TOOL_CALL_FAMILY,
+    QWEN_TOOL_CALL_FAMILY,
+    ChatModelMixin,
+)
 from .utils import get_context_length, get_max_src_len, pad_prefill_tokens
 
 logger = logging.getLogger(__name__)
@@ -62,6 +67,7 @@ NON_DEFAULT_MODEL_LIST: List[str] = [
     "MiniCPM-V-2.6",
     "glm-4v",
     "qwen2-vl-instruct",
+    "qwen2.5-vl-instruct",
     "qwen2-audio",
     "qwen2-audio-instruct",
     "deepseek-v2",
@@ -681,6 +687,8 @@ class PytorchChatModel(PytorchModel, ChatModelMixin):
             or model_family in LLAMA3_TOOL_CALL_FAMILY
         ):
             full_context_kwargs["tools"] = tools
+        elif tools and model_family in DEEPSEEK_TOOL_CALL_FAMILY:
+            self._tools_to_messages_for_deepseek(messages, tools)
         assert self.model_family.chat_template is not None
         full_prompt = self.get_full_context(
             messages,

xinference/model/llm/transformers/qwen2_audio.py

@@ -55,9 +55,9 @@ class Qwen2AudioChatModel(PytorchChatModel):
 
         device = self._pytorch_model_config.get("device", "auto")
         device = select_device(device)
-        self._device = device
         # for multiple GPU, set back to auto to make multiple devices work
         device = "auto" if device == "cuda" else device
+        self._device = device
 
         self._processor = AutoProcessor.from_pretrained(
             self.model_path,
@@ -105,6 +105,8 @@ class Qwen2AudioChatModel(PytorchChatModel):
         inputs = self._processor(
             text=text, audios=audios, return_tensors="pt", padding=True
         )
+        # Make sure that the inputs and the model are on the same device.
+        inputs.data = {k: v.to(self._device) for k, v in inputs.data.items()}
         inputs.input_ids = inputs.input_ids.to(self._device)
         generate_config = generate_config if generate_config else {}
         stream = generate_config.get("stream", False) if generate_config else False