Add lfm2.5 VL Export by NorbertKlockiewicz · Pull Request #17838 · pytorch/executorch

feat: add LFM2.5-VL-1.6B ExecuTorch export

NorbertKlockiewicz
and
claude
committed
fix(lfm2_5_vl): remove redundant re-export of text decoder

NorbertKlockiewicz
and
claude
committed
fix(lfm2_5_vl): use strict=False for token embedding export

NorbertKlockiewicz
and
claude
committed
fix(lfm2_5_vl): add full metadata set to match LLaMA/LFM2 text-only standard

NorbertKlockiewicz
and
claude
committed
feat(lfm2_5_vl): make max_seq_len/max_context_len configurable, derive metadata from ModelArgs

NorbertKlockiewicz
and
claude
committed
refactor(lfm2_5_vl): add config JSON, convert_weights, load params like LFM2 text-only

NorbertKlockiewicz
and
claude
committed
fix(lfm2_5_vl): catch Exception not KeyError in convert_weights fallback

NorbertKlockiewicz
and
claude
committed
feat(lfm2_5_vl): add BUCK file, activation stats logging, and smoke tests

NorbertKlockiewicz
and
claude
committed

Provide feedback