Behaviorally modified MXFP4 research checkpoint derived from XiaomiMiMo/MiMo-V2.6-Flash-RL.
Release status
| Item |
Status |
| Weight package |
MXFP4 mixed-precision safetensors included |
| Structural verification |
Complete |
| Native MiMo chat template |
Included |
| Baked deployment system prompt |
None |
| Refusal evaluation |
Pending |
| Comprehensive derivative benchmark |
Pending |
| Modification recipe |
Proprietary and intentionally not distributed |
| Access |
Public model page with manual approval gating |
DERISKED identifies the Blackfrost release family. It is not a claim of zero refusals, complete safety, harmlessness, or production readiness. Refusal measurements will be published only after evaluation of the exact uploaded artifact is complete.
Overview
MIMO-V2.6-DERISKED-MXFP4 is an independent Blackfrost research derivative of XiaomiMiMo/MiMo-V2.6-Flash-RL. The checkpoint retains the upstream MiMo V2.6 architecture, tokenizer, multimodal components, native chat template, and speculative-decoding assets while preserving its mixed-precision MXFP4 deployment layout.
The behavioral modification is encoded in the released weights. It is not a system-prompt wrapper, adapter, or decoding-time filter. Direction data, capture data, intermediate checkpoints, internal evaluation prompts, and the reproduction recipe are not included.
Model specifications
| Property |
Value |
| Architecture |
MiMoV2ForCausalLM / mimo_v2 |
| Parameters |
309B total / 15B activated, per the upstream card |
| Language-model layers |
48 |
| Hidden size |
4,096 |
| Routed experts |
256 total / 8 activated per token |
| Modalities |
Text, image, video, and audio |
| Upstream maximum context |
1,048,576 tokens |
| Current Blackfrost serving target |
262,144 tokens |
| Speculative decoder |
Five-layer MiMo MTP/DFlash assets retained |
| Main indexed tensor bytes |
172,923,364,096 |
| Expert deployment format |
Mixed-precision MXFP4 / ModelOpt layout |
The upstream architectural context limit is not a guarantee that every runtime or hardware configuration can allocate or serve the full window.
Lineage
XiaomiMiMo/MiMo-V2.6-Flash-RL
└── Blackfrost-AI/MIMO-V2.6-DERISKED-MXFP4
Tokenizer, architecture, multimodal, MTP/DFlash, and license lineage follow the upstream checkpoint. The language-model checkpoint contains Blackfrost weight-level behavioral modifications.
Prompting and chat template
The package uses the native MiMo chat template and supports role-structured messages, including a caller-supplied system role. No Blackfrost, Frosty, compliance, or deployment-specific system prompt is baked into this release.
Runtime reasoning and tool-call parsers must be compatible with MiMo. Applications remain responsible for tool execution, result reinjection, authentication, authorization, conversation state, and output handling.
Validation status
The release package has completed structural checks covering the indexed shard inventory, selected modified tensors, tensor shapes and dtypes, finite values, preserved MXFP4 expert shards, native configuration files, and tokenizer/template assets.
Refusal behavior, broad capability retention, coding, cybersecurity, multimodal operation, tool calling, speculative acceptance, and long-context behavior remain pending for the exact uploaded artifact. Upstream benchmark results must not be attributed to this modified checkpoint.
Access
This repository is public and manually gated. Request access through the Hugging Face model page. Approval grants access to the released files; it does not imply suitability for a particular workload or transfer responsibility for deployment controls to Blackfrost.
Limitations and responsibility
- This is an experimental research checkpoint.
- Generated content may be inaccurate, insecure, offensive, or otherwise unsuitable.
- The model is not a security boundary, policy engine, authorization mechanism, or substitute for professional judgment.
- Treat generated text, code, URLs, tool arguments, file paths, and commands as untrusted until independently reviewed.
- Operators are responsible for access controls, monitoring, legal compliance, and safeguards appropriate to their environment.
License and attribution
The upstream repository identifies the checkpoint license as MIT. Review the upstream model card and license metadata before use or redistribution.
Blackfrost is independent of and is not affiliated with, sponsored by, or endorsed by Xiaomi. This research artifact is provided as-is, without warranties.
Contact
For reproducible artifact issues, use this repository's Discussions. Do not post credentials, private prompts, personal information, or infrastructure details.
