Repository navigation

Update conversion for Luciole model (1B and 23B) - #1

Open

Jeronymous wants to merge 27 commits into

masterfrom

luciole

Open

Update conversion for Luciole model (1B and 23B)#1
Jeronymous wants to merge 27 commits into
masterfrom
luciole

Commits on Mar 18, 2026
- fix and update test
  Jeronymous
  committed
  d89fc07
- Update conversion script for Luciole
  Jeronymous
  committed
  ff6f474
Commits on Mar 25, 2026
- Fix tied word embeddings
  Jeronymous
  committed
  41c639c
Commits on Apr 1, 2026
- Special path for Luciole-8B (NemotronHForCausalLM hybrid archi)
  Jeronymous
  committed
  5c02601
- more robust distinction between Luciole and others
  Jeronymous
  committed
  c14603b
Commits on May 28, 2026
- Correctly handle added tokens that are not marked as special, such as <tool_call>. And fix conversion with tied weight embedding in some cases
  Jeronymous
  committed
  0284fd3
Commits on May 29, 2026
Commits on Jun 29, 2026
- test mutiturn
  Jeronymous
  committed
  ea85aa5
- use USER_DEFINED instead of CONTROL character. Add comments
  Jeronymous
  committed
  4ba0db6
- Test tokenizer with chat template when it is relevant
  Jeronymous
  committed
  666f106
- Do not add space prefix for Luciole, it can be more harmful than benefic (with chat template)
  Jeronymous
  committed
  333b8c1
Commits on Jul 15, 2026
- Add conversion in FP8 (transformer)
  Jeronymous
  committed
  feabe87
- Add Dockerfile
  Jeronymous
  committed
  ffe07d8
- Add helpers to run conversions and tests
  Jeronymous
  committed
  437a44c
- Add helpers to upload on Hugging Face
  Jeronymous
  committed
  4deb355
- Tune GPU memory utilization of vLLM to what is available. And limit max_model_len (to avoid useless OOM on small test)
  Jeronymous
  committed
  af79162
Commits on Jul 16, 2026
- particularities for mamba-transformer hybrid architecture (nemotron_h)
  Jeronymous
  committed
  9ad93a2
Commits on Jul 17, 2026
- Add a hack in NemotronH Modelfile so that it's not interpreted as a thinking model
  Jeronymous
  committed
  c5d5740
- implement thing for thinking models
  Jeronymous
  committed
  f183568
Commits on Aug 18, 2026
Commits on Sep 14, 2026
- Add conversion in FP4
  Jeronymous
  committed
  6595aca
Commits on Oct 14, 2026
- Add an option to skip GGUF conversion
  Jeronymous
  committed
  f0bb098
- Fix RoPE dimension for models with a custom head_dim != n_embd / n_head (see https://github.com/ggml-org/llama.cpp/issues/30518 and https://github.com/ggml-org/llama.cpp/pull/23343)
  Jeronymous
  committed
  f74fae2

Commits

Commits on Mar 18, 2026

Commits on Mar 25, 2026

Commits on Apr 1, 2026

Commits on May 28, 2026

Commits on May 29, 2026

Commits on Jun 29, 2026

Commits on Jul 15, 2026

Commits on Jul 16, 2026

Commits on Jul 17, 2026

Commits on Aug 18, 2026

Commits on Sep 14, 2026

Commits on Oct 14, 2026