lfm2a-fr

French fine-tunings of LFM2.5-Audio. Each model lives on its own branch: load it with LFM2AudioModel.from_pretrained(repo, revision=<branch>). The main branch only holds this index. Each branch's README gives the model's provenance and results.

Branch	Phase	Status	Parent	Trained on	Summary
s1-anchor-v1	s1-anchor	validé	LiquidAI/LFM2.5-Audio-1.5B	2026-08-27	Sequential French anchoring v1
s1-anchor-v2	s1-anchor	candidat	LiquidAI/LFM2.5-Audio-1.5B	2026-09-18	French anchoring v2 (sequential and interleaved)
s1-anchor-v3	s1-anchor	candidat	LiquidAI/LFM2.5-Audio-1.5B	2026-09-21	French anchoring v3 (all data, one epoch)
s2-ref	s2	candidat	LiquidAI/LFM2.5-Audio-1.5B	2026-09-25	French and English voice assistant with tool calling (phase 2 reference)
s2-tool-v2	s2	candidat	LiquidAI/LFM2.5-Audio-1.5B	2026-10-02	French and English voice assistant with tool calling (phase 2, tool-calling data v2)
s2-lora-p20	s2	candidat	LiquidAI/LFM2.5-Audio-1.5B	2026-10-09	French and English voice assistant with tool calling (phase 2 after a LoRA anchoring on 20 % of the anchoring corpus)
s2-lora-full	s2	candidat	LiquidAI/LFM2.5-Audio-1.5B	2026-10-10	French and English voice assistant with tool calling (phase 2 after a LoRA anchoring on a full epoch of the anchoring corpus)
s12-unique-sft	s12-unique	candidat	LiquidAI/LFM2.5-Audio-1.5B	2026-10-16	French and English voice assistant in one training phase (SFT on a balanced mix of ASR, TTS, replayed text, conversation and tool calling)
s12-unique-lora	s12-unique	candidat	LiquidAI/LFM2.5-Audio-1.5B	2026-10-17	French and English voice assistant in one training phase (LoRA on a balanced mix of ASR, TTS, replayed text, conversation and tool calling)
s13-tools-web-db-sft	s13-tools-web-db	candidat	LiquidAI/LFM2.5-Audio-1.5B	2026-10-27	French and English voice assistant with two general tools (SFT on the s12 mix plus query_database and web_search dialogues)