Code

Add weaviate vectors#31

Code

Merged

laugustyniak merged 61 commits intomasterpwr-ai/JuDDGES:masterfrom add-weaviate-vectorspwr-ai/JuDDGES:add-weaviate-vectorsCopy head branch name to clipboardDec 1, 2024

+558-362

Commits

Commits on Jun 24, 2024

• Add pre-commit and integrate it with CI

  binkjakub

  committed
  a3999bd

  Copy full SHA for a3999bd

• Apply pre-commit style fixes

  binkjakub

  committed
  8f189b3

  Copy full SHA for 8f189b3

• Add sorting imports to ruff

  binkjakub

  committed
  1b8e84e

  Copy full SHA for 1b8e84e

• Fix encoding in instruction dataset

  binkjakub

  committed
  96d6ca1

  Copy full SHA for 96d6ca1

Commits on Jul 3, 2024

• Refactor and make deterministic predictions

  binkjakub

  committed
  80f7002

  Copy full SHA for 80f7002

• Add llm-as-judge baseline results

  binkjakub

  committed
  04b8c47

  Copy full SHA for 04b8c47

• Make llm-as-judge using API instead of local LLM

  binkjakub

  committed
  4f79883

  Copy full SHA for 4f79883

Commits on Jul 4, 2024

• Fix missing attention_mask in prediction and add sampled predictions

  binkjakub

  committed
  54b9c8c

  Copy full SHA for 54b9c8c

• Update data description with output_size quantile

  binkjakub

  committed
  f85c48f

  Copy full SHA for f85c48f

• Update packages

  binkjakub

  committed
  1405116

  Copy full SHA for 1405116

Commits on Jul 8, 2024

• Refactor fine-tuning scripts and change hparams, make common code for unsloth and transformers

  binkjakub

  committed
  c9b7bf2

  Copy full SHA for c9b7bf2

• Refactor evaluation code

  binkjakub

  committed
  3a06f31

  Copy full SHA for 3a06f31

Commits on Jul 11, 2024

• Fix unsloth peft

  binkjakub

  committed
  ada17a3

  Copy full SHA for ada17a3

Commits on Jul 15, 2024

• Add optimized unsloth inference and inference telemetry

  binkjakub

  committed
  f2af0f7

  Copy full SHA for f2af0f7

Commits on Aug 8, 2024

• Reproduce SFT on update and fixed code

  binkjakub

  committed
  513343f

  Copy full SHA for 513343f

• Fix function evaluating info-extraction metrics

  binkjakub

  committed
  3c2cf3e

  Copy full SHA for 3c2cf3e

• Evaluate updated Mistral and summarize metrics with mean and std

  binkjakub

  committed
  4ca650e

  Copy full SHA for 4ca650e

Commits on Aug 9, 2024

• Fix computing maximum sequence length (account for output tokens during model's generation)

  binkjakub

  committed
  fadb582

  Copy full SHA for fadb582

Commits on Aug 11, 2024

• Add llm-as-judge preliminary results (too much non-evaluable)

  binkjakub

  committed
  1d8df86

  Copy full SHA for 1d8df86

• Add unit test for parsing prediction output

  binkjakub

  committed
  4161ba2

  Copy full SHA for 4161ba2

• Fix bug with CHRF computation

  binkjakub

  committed
  d056947

  Copy full SHA for d056947

• Add unit tests for chrf and fix metric params

  binkjakub

  committed
  00bffb3

  Copy full SHA for 00bffb3

• Rollback to legacy fine-tuning config due to unsloth compatibility issue

  binkjakub

  committed
  c46c476

  Copy full SHA for c46c476

Commits on Aug 12, 2024

• Add mistral-nemo LLM

  binkjakub

  committed
  ff2b3e3

  Copy full SHA for ff2b3e3

• Add cuda version argument in Makefile

  binkjakub

  committed
  f29864f

  Copy full SHA for f29864f

• Fix llm-as-judge implementation

  binkjakub

  committed
  7824769

  Copy full SHA for 7824769

• Disable CI on windows for now (utf-8 bugs)

  binkjakub

  committed
  d104981

  Copy full SHA for d104981

• Reproduce Mistral-Nemo and summarize all results

  binkjakub

  committed
  8ab3078

  Copy full SHA for 8ab3078

Commits on Aug 23, 2024

• Add prediction with OpenAI GPT models

  binkjakub

  committed
  f95ec86

  Copy full SHA for f95ec86

Commits on Aug 25, 2024

• Update README.md with reproduction instruction

  binkjakub

  committed
  57ea86d

  Copy full SHA for 57ea86d

• Migrate OpenAI interface to langchain to enable request caching

  binkjakub

  committed
  47a8f6b

  Copy full SHA for 47a8f6b

• Add results for gpt-4o and gpt-4o-mini

  binkjakub

  committed
  12c1bb9

  Copy full SHA for 12c1bb9

Commits on Aug 26, 2024

• Add Bielik v0.1 LLM

  Show description for ceaed8b
  binkjakub

  committed
  ceaed8b

  Copy full SHA for ceaed8b

• Add building english instruct dataset and infer over it with gpt-4o-mini

  binkjakub

  committed
  7dfd948

  Copy full SHA for 7dfd948

Commits on Aug 27, 2024

• Add optional chat templating

  binkjakub

  committed
  5665e75

  Copy full SHA for 5665e75

• Add new polish LLMs and separate configs for en and pl fine-tuned models

  binkjakub

  committed
  7d40fc5

  Copy full SHA for 7d40fc5

• Fix pl-court-raw dataset by filtering records with empty content from

  binkjakub

  committed
  5afd7e3

  Copy full SHA for 5afd7e3

Commits on Aug 28, 2024

• Add null-count info to pl-court-raw README

  binkjakub

  committed
  e9a1df5

  Copy full SHA for e9a1df5

• Reproduce Polish and English LLMs

  binkjakub

  committed
  bf0c034

  Copy full SHA for bf0c034

• Fix pre-commit exclude

  binkjakub

  committed
  6e18f71

  Copy full SHA for 6e18f71

• Add bielik v2

  binkjakub

  committed
  89f5199

  Copy full SHA for 89f5199

• Fix missing gpt-4o outputs and reproduce English data on gpt

  binkjakub

  committed
  d8b83c8

  Copy full SHA for d8b83c8

Commits on Aug 29, 2024

• Add llm-as-judge evaluation for English and Polish using gpt-4o-mini

  binkjakub

  committed
  f92a4e0

  Copy full SHA for f92a4e0

• Fix evaluation parser to account for non-yaml text at the beginning of the LLM output

  binkjakub

  committed
  593439d

  Copy full SHA for 593439d

• Reproduce evaluation on fixed parsing

  binkjakub

  committed
  a9cc9b3

  Copy full SHA for a9cc9b3

• Fix langchain versions

  binkjakub

  committed
  5f79cce

  Copy full SHA for 5f79cce

• Reproduce llm-as-judge with fixed prompt

  binkjakub

  committed
  e27516c

  Copy full SHA for e27516c

Commits on Aug 30, 2024

• Fix default num_proc in structured evaluator

  binkjakub

  committed
  3d2d38d

  Copy full SHA for 3d2d38d

Commits on Aug 31, 2024

• Reproduce Bielik v2

  binkjakub

  committed
  774048b

  Copy full SHA for 774048b

Commits on Sep 18, 2024

• Add text to chunk embeddings data and weaviate basic ingestion

  binkjakub

  committed
  12374a5

  Copy full SHA for 12374a5

• Add weaviate deployment

  binkjakub

  committed
  147815e

  Copy full SHA for 147815e

Commits on Sep 20, 2024

• Add two modes for weaviate ingest

  binkjakub

  committed
  4002989

  Copy full SHA for 4002989

• Add hf-transformers vectorizer to weaviate collection

  binkjakub

  committed
  ce619a7

  Copy full SHA for ce619a7

• Add weaviate example

  binkjakub

  committed
  40171ae

  Copy full SHA for 40171ae

• Add hf-transformer module to weaviate deployment

  binkjakub

  committed
  6fdcee4

  Copy full SHA for 6fdcee4

• Add weaviate deployment info

  binkjakub

  committed
  b305904

  Copy full SHA for b305904

• Add usage info to weaviate README.md

  binkjakub

  committed
  27c7eca

  Copy full SHA for 27c7eca

Commits on Sep 21, 2024

• Fix env variables naming

  binkjakub

  committed
  ef6170e

  Copy full SHA for ef6170e

Commits on Sep 23, 2024

• Fix env variables naming

  Show description for 13ac088
  binkjakub

  committed
  13ac088

  Copy full SHA for 13ac088

Commits on Oct 2, 2024

• Merge branch 'master' into add-weaviate-vectors

  Show description for 8510775
  binkjakub

  committed
  8510775

  Copy full SHA for 8510775

• Fix _moddix

  binkjakub

  committed
  e245236

  Copy full SHA for e245236