yangdx
0a693dbfda
Fix linting
2025-02-02 04:27:55 +08:00
yangdx
8484564f50
Fix llm_model_func retrieval error.
2025-02-02 03:54:41 +08:00
yangdx
873b52d2e4
Add debug logging for cache response retrieval
2025-02-02 03:15:43 +08:00
yangdx
bed5a97ae2
Fix prompt respond cache fail when is_embedding_cache_enabled is true
2025-02-02 03:09:06 +08:00
yangdx
5d14ab03eb
Fix linting
2025-02-02 01:56:32 +08:00
yangdx
b45ae1567c
Refactor LLM cache handling and entity extraction
...
- Removed custom LLM function in entity extraction
- Simplified cache handling logic
- Added `force_llm_cache` parameter
- Updated cache handling conditions
2025-02-02 01:28:46 +08:00
yangdx
6c7d7c25d3
Refactor cache handling logic for better readability, keep function unchanged.
2025-02-02 00:10:21 +08:00
yangdx
c9481c81b9
Add cache type "extract" for entity extraction
2025-02-01 23:05:02 +08:00
yangdx
3c3cdba499
Fix typo error
2025-02-01 22:27:49 +08:00
yangdx
3bc7c4d8f1
Save cache_type to llm_response_cache
2025-02-01 22:18:59 +08:00
yangdx
c3942077a9
Use direct embedding_func from hashing_kv (do not by pass maxiumu async control)
2025-02-01 22:12:45 +08:00
yangdx
c98a675b6c
remove unused parm
2025-02-01 22:07:12 +08:00
yangdx
95edf8a51e
Fix linting
2025-02-01 15:22:40 +08:00
yangdx
b109f57ddd
Refactor async call limiting to use asyncio.Semaphore for better performance.
...
- Replace custom counter with asyncio.Semaphore
- The existing implementation cannot follow the FIFO order
2025-02-01 10:36:25 +08:00
yangdx
2ba36f87e3
Add support for list input in quantize_embedding function
...
- Convert list to numpy array if needed
- Maintain existing functionality
2025-02-01 10:36:25 +08:00
yangdx
1192727be7
remove semaphore logic from EmbeddingFunc(cause num of instances is already control by limit_async_func_call)
2025-02-01 10:36:25 +08:00
yangdx
b0d87b2e29
Fix linting
2025-01-31 15:33:50 +08:00
yangdx
54b68074a1
Merge branch 'main' into fix-extract-entity-concurrent-problem
2025-01-31 01:19:44 +08:00
yangdx
3150c2812f
Fix loop step from 1 to 2 in get_conversation_turns, ensure user-assistant pairing
2025-01-30 13:08:27 +08:00
yangdx
cc50ade14e
Fix concurrent problem on extract_entities function.
...
- Abandon the approach of temporarily replacing the global llm_model_func configuration
- Introduce custom_llm function with new_config for handle_cache while extracting entities
- Update handle_cache to accept custom_llm
2025-01-30 02:45:33 +08:00
zrguo
80451af839
fix linting errors
2025-01-27 23:21:34 +08:00
Saifeddine ALOUI
16d1ae77ee
fixed csv_string_to_list when data contains null
2025-01-27 10:15:30 +01:00
Saifeddine ALOUI
315f0bf5f9
Added escaping to list_of_list_to_csv
2025-01-27 10:13:06 +01:00
Saifeddine ALOUI
06c9e4e454
Fixed missing imports bug and fixed linting
2025-01-25 00:55:07 +01:00
Saifeddine ALOUI
34018cb1e0
Separated llms from the main llm.py file and fixed some deprication bugs
2025-01-25 00:11:00 +01:00
Magic_yuan
5719aa8882
支持多轮对话
2025-01-24 19:03:36 +08:00
Magic_yuan
f6d29e1793
修复缓存处理的运行时错误
...
].func # ["func"]
^^^^
AttributeError: 'dict' object has no attribute 'func'
2025-01-24 11:01:32 +08:00
jin
6ae8647285
support pipeline mode
2025-01-16 12:58:15 +08:00
jin
d5ae6669ea
support pipeline mode
2025-01-16 12:52:37 +08:00
jin
85331e3fa2
update Oracle support
...
add cache support, fix bug
2025-01-10 11:36:28 +08:00
Samuel Chan
6ae27d8f06
Some enhancements:
...
- Enable the llm_cache storage to support get_by_mode_and_id, to improve the performance for using real KV server
- Provide an option for the developers to cache the LLM response when extracting entities for a document. Solving the paint point that sometimes the process failed, the processed chunks we need to call LLM again, money and time wasted. With the new option (by default not enabled) enabling, we can cache that result, can significantly save the time and money for beginners.
2025-01-06 12:50:05 +08:00
zrguo
a1493f56d6
Update utils.py
2024-12-26 22:09:46 +08:00
GG
04293466ad
fix(utils): 修复缓存未正确启用的问题
...
- 在 handle_cache 函数中添加了对 global_config 中 enable_l`lm_cache 设置的检查
- 如果配置禁用了缓存，则直接返回 None
-这个修改确保了在不需要缓存的情况下，函数能够正确地跳过缓存处理
2024-12-26 22:03:45 +08:00
david
f6eeedb050
add concurrent embedding limit
2024-12-10 09:01:21 +08:00
david
3210c8f5bd
fix unicode_escape.
2024-12-09 19:14:27 +08:00
partoneplay
a7fcb653e3
Merge remote-tracking branch 'origin/main' and fix syntax
2024-12-09 12:36:55 +08:00
Magic_yuan
ccf44dc334
feat(cache): 增加 LLM 相似性检查功能并优化缓存机制
...
- 在 embedding 缓存配置中添加 use_llm_check 参数
- 实现 LLM 相似性检查逻辑，作为缓存命中的二次验证- 优化 naive 模式的缓存处理流程
- 调整缓存数据结构，移除不必要的 model 字段
2024-12-08 17:35:52 +08:00
Magic_yuan
dc2a45004a
缓存计算函数迁移到工具类
2024-12-08 10:37:55 +08:00
magicyuan876
786aa92b48
fix(utils): 修复 JSON 日志编码问题
...
- 在 json.dumps 中添加 ensure_ascii=False 参数，以支持非 ASCII 字符编码
-这个修改确保了包含中文等非 ASCII 字符的日志信息能够正确处理和显示
2024-12-06 14:32:41 +08:00
magicyuan876
e619b09c8a
重构缓存处理逻辑
...
- 提取通用缓存处理逻辑到新函数 handle_cache 和 save_to_cache
- 使用 CacheData 类统一缓存数据结构
- 优化嵌入式缓存和常规缓存的处理流程
- 添加模式参数以支持不同查询模式的缓存策略
- 重构 get_best_cached_response 函数，提高缓存查询效率
2024-12-06 14:29:16 +08:00
magicyuan876
2ecdab2f18
整理了一下手误的注释
2024-12-06 09:11:22 +08:00
magicyuan876
d48c6e4588
feat(lightrag): 添加 查询时使用embedding缓存功能
...
- 在 LightRAG 类中添加 embedding_cache_config配置项
- 实现基于 embedding 相似度的缓存查询和存储
- 添加量化和反量化函数，用于压缩 embedding 数据
- 新增示例演示 embedding 缓存的使用
2024-12-06 08:17:20 +08:00
b10902118
753c1e6714
support JSON output for ollama and openai
2024-11-29 21:41:37 +08:00
jin
26ae240c65
Logic Optimization
2024-11-25 13:40:38 +08:00
jin
1dbe803521
Merge branch 'main' of https://github.com/jin38324/LightRAG
2024-11-25 13:32:33 +08:00
jin
89c2de54a2
Optimization logic
2024-11-25 13:29:55 +08:00
jin
662303f605
use oracle bind variables to avoid error
2024-11-15 12:57:01 +08:00
LarFii
f8b15df74c
fix cache bug
2024-11-14 15:59:37 +08:00
Ken Wiltshire
3d5d083f42
fix event loop conflict
2024-11-06 11:18:14 -05:00
zrguo
3f7ae11962
Update utils.py
2024-11-03 17:53:53 +08:00
