staticlabs/dlm-code0.6b-exp

1---2tags:3- masked-diffusion4- diffusion5- qwen26- code7language:8- en9pipeline_tag: text-generation10---11 12# dlm-code0.6b-exp13 14**Experimental** masked-diffusion language model for code, from the [thunder-fast](https://github.com/AdrianTuci1/thunder-fast) project.15 16This is a **discrete masked-diffusion (MDM)** adaptation of a Qwen2-based pretrained code model (~0.5B). It is *not* an autoregressive next-token model: generation is a progressive un-masking loop under **bidirectional attention**, so a whole output window is filled in parallel over 24 diffusion steps.17 18## Details19 20| | |21|---|---|22| Backbone | Qwen2 (`model_type: qwen2`), 24 layers, hidden 896, 14 attention / 2 KV heads, GQA |23| Vocabulary | 151,936 + `<M>` mask token (id 151665) |24| Diffusion | discrete `[MASK]`, 24-step progressive un-masking, entropy-based position confidence (`alg_temp` 0.6, `top_k` 500) |25| Context | window-based (256 tokens) — not left-to-right; supports infilling |26| Weights | bf16 (`model.safetensors`) |27 28## How to run29 30This checkpoint is intended to be driven by a **diffusion decoder** (bidirectional attention + the iterative un-masking loop), e.g. the thunder-fast runtime. Loading it as a plain causal `Qwen2ForCausalLM` will *not* reproduce diffusion generation.31 32```33# thunder-fast runtime (ggml engine) or the reference DiffusionLM34# e.g. runtime --model dlm-code0.6b-exp.gguf --prompt "..."35```36 37## Status38 39Experimental / work-in-progress (`.exp`). It is a diffusion code-generation checkpoint published for the thunder-fast project; see the repo for training, the masking schedule, and evaluation details.40