The model card, file list, and checksums are here. Have the files and permission to share them? Submit a torrent so others can download from peers.
Getting the source files requires Hugging Face approval.Drop-in abliterated weights for XiaomiMiMo/MiMo-V2.6-Flash-RL. Same layout as the base checkpoint: FP8 attention, MXFP4 experts, DFlash draft head, and MTP.
| Base | XiaomiMiMo/MiMo-V2.6-Flash-RL |
| License | MIT, inherited from the base |
| Refusal suite | 27/32 BYPASS · 5 refuse · 0 garble |
| Cyber suite | 22/22 BYPASS · 0 refuse · 0 garble |
| Edit | Decoder self_attn.o_proj only · λ=3.5 |
| Left stock | Experts, vision, audio, embeddings, DFlash head, MTP |
Thinking was off for both suites. A reply counts only when it starts delivering the requested content. This is not a 32/32 checkpoint.
Full credit: XiaomiMiMo/MiMo-V2.6-Flash-RL · tonyd2wild/MiMo-V2.6-Flash-2x-DGX-Spark
This model has had safety refusals removed. That makes it useful for red-teaming, security research, evaluation, and unfiltered assistant tasks, and it removes guardrails you must supply yourself.
Access is gated. Agreeing to the terms grants access automatically. By requesting access, downloading, or using these weights, you agree to the terms below.
See RESPONSIBLE_USE.md.
Only decoder attention output projections (model.layers.*.self_attn.o_proj) differ from the base. λ is 3.5. Experts, the vision encoder, the audio encoder, embeddings, the DFlash draft under dflash/, and model_mtp.safetensors are the base files.
DFlash reads target layers 0, 11, 23, 35, and 47. Layers 0, 11, 23, and 35 were not edited. Layer 47 o_proj has a surgical row edit, so a stock DFlash head can accept fewer draft tokens than it does on the unmodified base. The target model still produces the abliterated continuation.
On DGX Spark (GB10), keep --gpu-memory-utilization and SGLang --mem-fraction-static at or below 0.85.
The measured suites were collected from an SGLang EAGLE serve, thinking off. The same weights load in the vLLM DFlash recipe at tonyd2wild/MiMo-V2.6-Flash-2x-DGX-Spark. That recipe's dflash/config.json trailing-comma fix still applies. Pass chat_template_kwargs.enable_thinking: false when you want the thinking-off behavior measured above.
Vendor sampling defaults are temperature 1.0 and top_p 0.95.
model_pp0_ep0_shard0.safetensors 13 GBdflash/dflash_draft_model.safetensors 2.9 GBmodel_pp0_ep25_shard0.safetensors 2.5 GBmodel_pp0_ep26_shard0.safetensors 2.5 GBmodel_pp0_ep27_shard0.safetensors 2.5 GBmodel_pp0_ep28_shard0.safetensors 2.5 GBmodel_pp0_ep29_shard0.safetensors 2.5 GBmodel_pp0_ep30_shard0.safetensors 2.5 GBmodel_pp0_ep31_shard0.safetensors 2.5 GBmodel_pp0_ep32_shard0.safetensors 2.5 GBmodel_pp0_ep33_shard0.safetensors 2.5 GBmodel_pp0_ep34_shard0.safetensors 2.5 GBmodel_pp0_ep35_shard0.safetensors 2.5 GBmodel_pp0_ep36_shard0.safetensors 2.5 GBmodel_pp0_ep37_shard0.safetensors 2.5 GBmodel_pp0_ep38_shard0.safetensors 2.5 GBmodel_pp0_ep39_shard0.safetensors 2.5 GBmodel_pp0_ep40_shard0.safetensors 2.5 GBmodel_pp0_ep41_shard0.safetensors 2.5 GBmodel_pp0_ep42_shard0.safetensors 2.5 GBmodel_pp0_ep43_shard0.safetensors 2.5 GBmodel_pp0_ep44_shard0.safetensors 2.5 GBmodel_pp0_ep45_shard0.safetensors 2.5 GBmodel_pp0_ep46_shard0.safetensors 2.5 GBmodel_pp0_ep47_shard0.safetensors 2.5 GBmodel_pp0_ep48_shard0.safetensors 2.5 GBmodel_pp0_ep49_shard0.safetensors 2.5 GBmodel_pp0_ep50_shard0.safetensors 2.5 GBmodel_pp0_ep51_shard0.safetensors 2.5 GBmodel_pp0_ep52_shard0.safetensors 2.5 GBmodel_pp0_ep53_shard0.safetensors 2.5 GBmodel_pp0_ep54_shard0.safetensors 2.5 GBmodel_pp0_ep55_shard0.safetensors 2.5 GBmodel_pp0_ep56_shard0.safetensors 2.5 GBmodel_pp0_ep57_shard0.safetensors 2.5 GBmodel_pp0_ep58_shard0.safetensors 2.5 GBmodel_pp0_ep59_shard0.safetensors 2.5 GBmodel_pp0_ep60_shard0.safetensors 2.5 GBmodel_pp0_ep61_shard0.safetensors 2.5 GBmodel_pp0_ep62_shard0.safetensors 2.5 GBmodel_pp0_ep63_shard0.safetensors 2.5 GBmodel_pp0_ep10_shard0.safetensors 2.5 GBmodel_pp0_ep11_shard0.safetensors 2.5 GBmodel_pp0_ep12_shard0.safetensors 2.5 GBmodel_pp0_ep13_shard0.safetensors 2.5 GBmodel_pp0_ep14_shard0.safetensors 2.5 GBmodel_pp0_ep15_shard0.safetensors 2.5 GBmodel_pp0_ep16_shard0.safetensors 2.5 GBmodel_pp0_ep17_shard0.safetensors 2.5 GBmodel_pp0_ep18_shard0.safetensors 2.5 GBmodel_pp0_ep19_shard0.safetensors 2.5 GBmodel_pp0_ep20_shard0.safetensors 2.5 GBmodel_pp0_ep21_shard0.safetensors 2.5 GBmodel_pp0_ep22_shard0.safetensors 2.5 GBmodel_pp0_ep23_shard0.safetensors 2.5 GBmodel_pp0_ep24_shard0.safetensors 2.5 GBmodel_pp0_ep3_shard0.safetensors 2.5 GBmodel_pp0_ep4_shard0.safetensors 2.5 GBmodel_pp0_ep5_shard0.safetensors 2.5 GBmodel_pp0_ep6_shard0.safetensors 2.5 GBmodel_pp0_ep7_shard0.safetensors 2.5 GBmodel_pp0_ep8_shard0.safetensors 2.5 GBmodel_pp0_ep9_shard0.safetensors 2.5 GBmodel_pp0_ep2_shard0.safetensors 2.5 GBmodel_pp0_ep1_shard0.safetensors 2.5 GBaudio_tokenizer/model.safetensors 1.9 GBmodel_mtp.safetensors 1.2 GBdflash/mask_embedding.pt 9.9 kBXiaomiMiMo/MiMo-V2.6-Flash-RL