[pull] main from abetlen:main

CHANGELOG.md

@@ -7,6 +7,11 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [0.3.13]
+
+- feat: Update llama.cpp to ggerganov/llama.cpp@
+- fix: Better chat format for Qwen2.5-VL by @alcoftTAO in #2040
+
 ## [0.3.12]
 
 - feat: Update llama.cpp to ggerganov/llama.cpp@a0374a67e2924f2e845cdc59dd67d9a44065a89c

llama_cpp/__init__.py

@@ -1,4 +1,4 @@
 from .llama_cpp import *
 from .llama import *
 
-__version__ = "0.3.12"
+__version__ = "0.3.13"

llama_cpp/llama_chat_format.py

@@ -3459,26 +3459,29 @@ class Qwen25VLChatHandler(Llava15ChatHandler):
     DEFAULT_SYSTEM_MESSAGE = "You are a helpful assistant."
 
     CHAT_FORMAT = (
-        "<|im_start|>system\n"
-        "You are a helpful assistant.<|im_end|>\n"
+        #"{% set image_count = namespace(value=0) %}"
+        #"{% set video_count = namespace(value=0) %}"
         "{% for message in messages %}"
-        "{% if message['role'] == 'user' %}"
-        "<|im_start|>user\n"
+        "{% if loop.first and message['role'] != 'system' %}"
+        "<|im_start|>system\n"
+        "{{ self.DEFAULT_SYSTEM_MESSAGE }}<|im_end|>\n"
+        "{% endif %}"
+        "<|im_start|>{{ message['role'] }}\n"
         "{% if message['content'] is string %}"
-        "{{ message['content'] }}"
+        "{{ message['content'] }}<|im_end|>\n"
         "{% else %}"
         "{% for content in message['content'] %}"
-        "{% if content['type'] == 'text' %}"
-        "{{ content['text'] }}"
-        "{% elif content['type'] == 'image_url' %}"
+        "{% if content['type'] == 'image_url' %}"
         "{% if content.image_url is string %}"
         "{{ content.image_url }}"
         "{% else %}"
         "{{ content.image_url.url }}"
         "{% endif %}"
+        #"{% set image_count.value = image_count.value + 1 %}"
+        "{% elif content['type'] == 'text' %}"
+        "{{ content['text'] }}"
         "{% endif %}"
         "{% endfor %}"
-        "{% endif %}"
         "<|im_end|>\n"
         "{% endif %}"
         "{% endfor %}"

llama_cpp/llama_cpp.py

@@ -200,6 +200,7 @@
 """RWKV tokenizer based on greedy tokenization"""
 
 
+# NOTE: Deprecated and will be removed in the future. (already gone in llama.cpp)
 # // pre-tokenization types
 # enum llama_vocab_pre_type {
 #     LLAMA_VOCAB_PRE_TYPE_DEFAULT        = 0,