28ac031ff6
feat: use llamaindex everywhere ( #525 )
...
* feat: use llamaindex for transcript final title too
* refactor: removed llm backend, replaced with one single class+llamaindex
* refactor: self-review
* fix: typing
* fix: tests
* refactor: extract clean_title and add tests
* test: fix
* test: remove ensure_casing/nltk
* fix: tiny mistake
2025-08-01 12:13:00 -06:00
ad56165b54
fix: remove unused settings and utils files ( #522 )
...
* fix: remove unused settings and utils files
* fix: remove migration done
* fix: remove outdated scripts
* fix: removing deployment of hermes, not used anymore
* fix: partially remove secret, still have to understand frontend.
2025-07-31 17:45:48 -06:00
406164033d
feat: new summary using phi-4 and llama-index ( #519 )
...
* feat: add litellm backend implementation
* refactor: improve generate/completion methods for base LLM
* refactor: remove tokenizer logic
* style: apply code formatting
* fix: remove hallucinations from LLM responses
* refactor: comprehensive LLM and summarization rework
* chore: remove debug code
* feat: add structured output support to LiteLLM
* refactor: apply self-review improvements
* docs: add model structured output comments
* docs: update model structured output comments
* style: apply linting and formatting fixes
* fix: resolve type logic bug
* refactor: apply PR review feedback
* refactor: apply additional PR review feedback
* refactor: apply final PR review feedback
* fix: improve schema passing for LLMs without structured output
* feat: add PR comments and logger improvements
* docs: update README and add HTTP logging
* feat: improve HTTP logging
* feat: add summary chunking functionality
* fix: resolve title generation runtime issues
* refactor: apply self-review improvements
* style: apply linting and formatting
* feat: implement LiteLLM class structure
* style: apply linting and formatting fixes
* docs: env template model name fix
* chore: remove older litellm class
* chore: format
* refactor: simplify OpenAILLM
* refactor: OpenAILLM tokenizer
* refactor: self-review
* refactor: self-review
* refactor: self-review
* chore: format
* chore: remove LLM_USE_STRUCTURED_OUTPUT from envs
* chore: roll back migration lint changes
* chore: roll back migration lint changes
* fix: make summary llm configuration optional for the tests
* fix: missing f-string
* fix: tweak the prompt for summary title
* feat: try llamaindex for summarization
* fix: complete refactor of summary builder using llamaindex and structured output when possible
* fix: separate prompt as constant
* fix: typings
* fix: enhance prompt to prevent mentioning others subject while summarize one
* fix: various changes after self-review
* fix: from igor review
---------
Co-authored-by: Igor Loskutov <igor.loskutoff@gmail.com >
2025-07-31 15:29:29 -06:00
cfb1b2f9bc
Upgrade modal apps
2025-03-25 11:09:01 +01:00
163d4a6e4a
Refactor transcribe segment
2025-01-20 12:46:20 +01:00
99ff06ff17
OpenAI compatible transcription api
2025-01-20 12:27:58 +01:00
7ff201f3ff
Fix model download
2024-12-27 14:23:03 +01:00
895ba36cb9
fix: modal upgrade ( #421 )
2024-10-01 16:39:24 +02:00
5267ab2d37
feat: retake summary using NousResearch/Hermes-3-Llama-3.1-8B model ( #415 )
...
This feature a new modal endpoint, and a complete new way to build the
summary.
## SummaryBuilder
The summary builder is based on conversational model, where an exchange
between the model and the user is made. This allow more context
inclusion and a better respect of the rules.
It requires an endpoint with OpenAI-like completions endpoint
(/v1/chat/completions)
## vLLM Hermes3
Unlike previous deployment, this one use vLLM, which gives OpenAI-like
completions endpoint out of the box. It could also handle guided JSON
generation, so jsonformer is not needed. But, the model is quite good to
follow JSON schema if asked in the prompt.
## Conversion of long/short into summary builder
The builder is identifying participants, find key subjects, get a
summary for each, then get a quick recap.
The quick recap is used as a short_summary, while the markdown including
the quick recap + key subjects + summaries are used for the
long_summary.
This is why the nextjs component has to be updated, to correctly style
h1 and keep the new line of the markdown.
2024-09-14 02:28:38 +02:00
Sara
004787c055
upgrade modal
2024-08-12 12:24:14 +02:00
projects-g
06b0abaf62
deployment fix ( #364 )
2024-06-20 12:07:28 +05:30
