Skip to content

Release 0.2.0#25

Draft
Popochounet wants to merge 72 commits into
mainfrom
dev
Draft

Release 0.2.0#25
Popochounet wants to merge 72 commits into
mainfrom
dev

Conversation

@Popochounet

Copy link
Copy Markdown
Contributor
  • MAJOR: Complete refacto with Ray and k8s

  • Client refacto: websocket connection. Client can send Audio & Text.

  • Complete config files for HuRI, k8s, Clients, Modules

  • New modules:

  1. MIC: vad mic that detect speech
  2. STT: speech to text that transcribe speech with sliding windows
  3. TAG: transcript aggregator to aggregate transcript into one sentence when end of speech
  4. RAG: Call RAGHandle (Ollama, Qdrant, embedding, history) to get answer as stream
  5. TTS: answer to audio speech
  6. MOV: gesture generation

Popochounet and others added 30 commits May 4, 2026 07:07
* wip(Embedding): pseudo module

* evol(events): error handling

* evol(MIC): WebRTC vad to detect if speech, no longer a threshold

* evol(STT): use of faster whisper can transcript in real time + sliding window transcription with end boolean

* evol(client config): sample rate, frame duration + mic,stt modules args

* evol(sender): now can send json serializable object only

* fix(client config): wrong arg name

* feat(TAG): aggregate all transcription into one question

* fix(linter): make lint

* fix(TAG): some match would destroy sentence

* evol(requirement): new modules

* evol(config): useless config + handle

* evol(README): launch_huri without config file

* fix(connection): handle errors from client - server connection

---------

Co-authored-by: Popochounet <adrien.audiard@epitech.eu>
* wip(rag): V1 of a rag working with an ollama llm working with the current pipeline with a RAG + Embedding + LLM (ollama). Should work with vLLM but not tested

* wip(rag): set the filter at None to be able to restrieve collections without a user_id

* feat(id): add ids to make it work with the rag system + an ingestion system

* clean(id): clean code

* wip(pr): add a module with ids and generate a rag class with module with id and module with handle. Make some refacto : user_id -> _user_id and handle -> _handle
* wip(rag): V1 of a rag working with an ollama llm working with the current pipeline with a RAG + Embedding + LLM (ollama). Should work with vLLM but not tested

* wip(rag): set the filter at None to be able to restrieve collections without a user_id

* feat(id): add ids to make it work with the rag system + an ingestion system

* clean(id): clean code

* feat(ingestion): ingestion done with the possibility of semantic and word base ingestion.

* wip(todo): Add some todos to not forget the work I have to do

* refacto(user_ids): user_id -> user_id

* wip(pr): add a module with ids and generate a rag class with module with id and module with handle. Make some refacto : user_id -> _user_id and handle -> _handle

* wip(docker): add a docker that launch with one commande. Only work with Ollama for other provider we need to change the code.

* wip(config): move qdrant, ollama into a config file.

* wip(config): config file client updated

* feat(huri): update config file

* fix(ingestion): update for the wrong branch now fixed

* merge: dev -> launch docker

* fix(config): huri.yaml fix

* remove(main): remove unnecessary function

* wip(lint): cleaner code with a make lint

* update(requierements): update requierements.txt

* delete(reasoning): Reasoning for later
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants