241 Commits

Author SHA1 Message Date
Tim Jaeryang Baek 039f9c54f8 Merge pull request #544 from minijeong-log/main
feat: add Langfuse v3 filter pipeline
2025-08-19 03:05:01 +04:00
Tim Jaeryang Baek a2e74ef9e5 Merge pull request #546 from kikumoto/feature/add_support_opus-4.1
Support Anthropic Claude 4.1; avoid simultaneous temperature and top_p
2025-08-19 03:04:37 +04:00
Olv Grolle 49c0494c66 Update Google Vertex AI pipeline with enhanced image support and latest models
- Add co-author credit
- Import base64 for image processing
- Update model list with Gemini 2.5 models and reorganize by version
- Improve image handling with proper MIME type detection and validation
- Add comprehensive debug logging for message processing
- Fix image data processing to use Part.from_data with proper MIME types
- Add error handling for unsupported image formats
2025-08-14 16:53:51 +02:00
Takahiro Kikumoto ce82635225 feat: Support Claude 4.1; avoid simultaneous temperature and top_p
Primary goal: enable compatibility with Anthropic Claude 4.1. Claude 4.1 rejects requests that specify both temperature and top_p. Align with API guidance and use only one: default to temperature; if top_p is explicitly provided, omit temperature. Preserve existing system/stop behavior. Apply equivalent handling for AWS Bedrock Claude (topP) for parity. Thinking-mode overrides remain unchanged (temperature=1.0; no top_p/topP).
2025-08-08 22:54:51 +09:00
mini.jeong cc79a91325 feat: add langfuse v3 filter
Author: Mini Jeong <mini.jeong@navercorp.com>
2025-07-31 14:18:21 +09:00
ther3zz bda1c8ef65 Update langfuse_filter_pipeline.py
update listed version to 1.7.1
2025-07-09 15:28:12 -04:00
ther3zz 0e8b730d87 change required langfuse version
set langfuse<3.0.0
2025-07-09 15:24:33 -04:00
Tim Jaeryang Baek f86f1475a4 Merge pull request #522 from kikumoto/feature/add_suuport_thinking_for_sonnet4_opus4
feat: Add Claude-4 thinking support
2025-07-08 22:32:49 +04:00
Chris Tran 7d0c33b033 Use session_id for tracing temporary chats with Langfuse 2025-06-13 13:02:03 -06:00
Takahiro Kikumoto 27353f7402 feat: Add Claude-4 thinking support for Bedrock 2025-05-31 17:04:48 +09:00
Takahiro Kikumoto 25ddc603e8 feat: Add Claude-4 thinking support
- Support thinking for claude-sonnet-4 and claude-opus-4
- Add get_thinking_supported_models() for better maintainability
2025-05-31 16:49:37 +09:00
Thomas DELORGE ba663e263c 🆕[Anthropic] Add new Claude 4 Sonnet & Opus 2025-05-22 20:10:26 +02:00
Lorenzo 72768c272e fix: removed invalid syntax 2025-04-18 21:47:07 +00:00
Lorenzo f825066804 Merge branch 'open-webui:main' into main 2025-04-18 18:40:36 -03:00
Lorenzo e9865901dd new: filter logic for a better way of filtering models 2025-04-18 21:39:24 +00:00
Tim Jaeryang Baek 6bcd91519a Merge pull request #466 from Lothiraldan/opik-filter-example
Add filter example of message monitoring using Opik
2025-04-14 08:56:03 -07:00
Tim Jaeryang Baek ef900c4a3b Merge pull request #470 from kikumoto/feature/update_aws_bedrock_claude_implementation
Feature/update aws bedrock claude implementation
2025-04-14 08:55:42 -07:00
Tim Jaeryang Baek 419779101c Merge pull request #469 from kikumoto/feature/support_for_bedrock_deepseek
Add  AWS Bedrock DeepSeek model.
2025-04-14 08:55:32 -07:00
Tim Jaeryang Baek 99cfc6bdb9 Merge pull request #467 from bartonzzx/langgraph-integration
Add langgraph integration example, also support thinking.
2025-04-14 08:55:21 -07:00
Tim Jaeryang Baek 6c6e3bd45d Merge pull request #460 from reasv/patch-1
Fix: Error: int() argument must be a string, a bytes-like object or a…
2025-04-14 08:54:38 -07:00
Tim Jaeryang Baek b82442e03e Merge pull request #479 from angkk2u/main
Update perplexity new API models
2025-04-12 14:10:18 -07:00
Tim Jaeryang Baek 67699acde9 Merge pull request #489 from Eisaichen/main
Update google_manifold_pipeline.py
2025-04-12 14:10:09 -07:00
Eric Z af479bc7ce fix(flowise): minor cleanup and finalize of integration hook 2025-04-12 15:14:59 -05:00
Eric Z cb5a16a5a4 fix(flowise): add streaming capability to flowise integration 2025-04-12 15:14:52 -05:00
Eric Z 5e1f90dc49 feat(flowise): add flowise integration for agentic evaluation 2025-04-12 15:14:44 -05:00
Timothy Jaeryang Baek cf57fd0747 Create events_pipeline.py
Co-Authored-By: Anthony Durussel <87324020+anthonydurussel@users.noreply.github.com>
2025-04-10 09:24:11 -07:00
Eisai eb9a3a2d6d Update google_manifold_pipeline.py 2025-04-04 23:24:20 -05:00
Timothy Jaeryang Baek 275655fd2e Merge pull request #476 from NolanTrem/main
Add R2R Example Pipeline
2025-03-29 15:12:19 -07:00
Timothy Jaeryang Baek 5ccdecc4e3 Merge pull request #438 from ther3zz/patch-3
langfuse pipeline: chatID + clearer observation names + usage + trace tagging
2025-03-28 15:37:00 -07:00
Timothy Jaeryang Baek 86bceb126f Merge pull request #473 from xxond/patch-1
Update Gemini Pipelines to 2.0 in examples
2025-03-28 13:12:19 -07:00
ther3zz c0a60f1b0f Model Name vs ID for Generation
adds the ability to use the model name as the model value when inserting generation observations into langfuse.

Also adds both model name and id in metadata.
2025-03-28 10:30:32 -04:00
Erik 4b42a7b514 Create azure_dalle_manifold_pipeline.py
A pipeline for generating text and processing images using the Azure API. And including multiple Dall-e models
2025-03-26 08:02:25 +01:00
angkk2u 83716778a6 Update new API models
Updates based on the new API models
2025-03-26 12:55:08 +09:00
NolanTrem ece9fceb59 Add R2R Example Pipeline 2025-03-21 16:53:19 -07:00
xxond c04d6f41ce Update pipeline list in google_vertexai_manifold_pipeline.py 2025-03-20 19:31:12 +03:00
Takahiro Kikumoto 488c43edd9 improve: enhance image format detection in process_image method
- Add proper MIME type detection for both data URLs and HTTP requests
- Extract media type from Content-Type header or MIME type
- Make format detection more robust and generic
- Remove hardcoded PNG/JPEG format assumptions
2025-03-18 21:28:05 +09:00
Takahiro Kikumoto 327062733a Add support for Claude 3.7 thinking mode
- Implemented support for Claude 3.7 thinking mode by adding reasoning effort and budget tokens.
- Added checks to ensure combined tokens do not exceed the maximum limit.
- Adjusted inference configuration to accommodate thinking mode requirements.
- Referenced implementation from https://github.com/open-webui/pipelines/blob/main/examples/pipelines/providers/anthropic_manifold_pipeline.py.
2025-03-18 17:38:46 +09:00
Takahiro Kikumoto ecc44ebd1e Enhance get_models method to include models with INFERENCE_PROFILE type
- Updated the get_models method to fetch models that support both ON_DEMAND and INFERENCE_PROFILE inference types.
- Added a helper method getInferenceProfileId to retrieve the inference profile ID for models with INFERENCE_PROFILE type.
- This change ensures that models with different inference types are correctly listed and available for use.
2025-03-18 16:48:12 +09:00
Takahiro Kikumoto c1bbbe1165 Refactor AWS Bedrock Claude Pipeline to support Instance Profile and Task Role
- Updated `Valves` class to use `Optional[str]` for AWS credentials.
- Modified `__init__` method to initialize `valves` with environment variables.
- Added `update_pipelines` method to handle Bedrock client initialization and model fetching.
- Refactored `on_startup` and `on_valves_updated` methods to call `update_pipelines`.
- Improved error handling in `update_pipelines` and `get_models` methods.
2025-03-18 16:33:44 +09:00
Takahiro Kikumoto 51e267c10f Refactor payload structure to comply with Bedrock Converse API
- Updated `inferenceConfig` to include `temperature`, `topP`, `maxTokens`, and `stopSequences`.
- Added `additionalModelRequestFields` with `top_k` parameter.
- Removed unnecessary deletion of `additionalModelRequestFields` in `stream_response` method.
2025-03-18 13:29:15 +09:00
Takahiro Kikumoto 827b47d2d5 Fix system message handling and payload cleanup in AWS Bedrock Claude Pipeline
- Corrected the system message extraction to use the "content" field.
- Removed unnecessary deletion of the "system" field from the payload in stream_response method.
2025-03-18 12:01:14 +09:00
Takahiro Kikumoto c5af48c45b Add support for outputting reasoning process in DeepSeek
- Modified `stream_response` method to handle reasoning content blocks.
- Added logic to yield "<think>" and "</think>" tags around reasoning content.
2025-03-18 10:34:17 +09:00
Takahiro Kikumoto 9ec364092d Implemented pipeline for AWS Bedrock DeepSeek model. 2025-03-17 17:54:27 +09:00
bartonzzx 48ddbec455 Add langgraph integration example, also support thinking. 2025-03-15 23:02:38 +08:00
Boris Feld b7debc3b1b Add filter example of message monitoring using Opik 2025-03-13 17:46:51 +01:00
reasv afb4704034 Fix: Error: int() argument must be a string, a bytes-like object or a real number, not 'NoneType' 2025-03-10 23:22:43 +01:00
ther3zz 34a464e1d9 fix model id
we now pass model info to langfuse for generation observations
2025-03-04 13:27:58 -05:00
ther3zz 40c32238af show debug switch last
moving the debug valve as the last option (above the tags valve)
2025-03-04 10:32:40 -05:00
ther3zz 3703976ac2 modify non-llm responses as events
instead of inserting non-llm responses as generation, they are now inserted as events into langfuse
2025-03-04 10:28:20 -05:00
ther3zz 689d986f81 add tag valve and tag insert
adds a valve to enable/disable inserting tags into the langfuse trace.
2025-03-03 13:26:03 -05:00