| license: other | |
| tags: | |
| - dflash | |
| - speculative-decoding | |
| - laguna | |
| base_model: | |
| - poolside/Laguna-S-2.1 | |
| # Laguna-S-2.1-DFlash-polishing | |
| DFlash speculator (drafter) for Laguna-S-2.1, bf16. | |
| - Architecture: `DFlashLagunaForCausalLM` (6 sliding-attention layers, block_size 16). | |
| - Shares token embedding + lm_head with the target; `draft_vocab_size == vocab_size` (no d2t/t2d). | |
| - Source checkpoint: `s3://poolside.titan.checkpoints.us-east-2/adam/checkpoints/ft_sft_e0630_rhiemann_baseline_titan_sft_training/0008400/DFlash_Stage2/25000` | |
| - Loads under vLLM (native `laguna_dflash`) and TRT-LLM (pytorch DFlash backend) as the draft model in a speculative config. | |