MiniMax M3.1 Flash DFlash2 draft (incoai), mirror

Mirror of incoai/MiniMax-M3.1-Flash-DFlash2@ac07a7c1ecaae26a7ace595756124dca974b432d for reproducibility of github.com/hyukudan/strix-halo-engine-lab.

The files are exact copies of that upstream revision: same paths, same bytes. They were not re-quantized or modified. All credit belongs to the original authors below; please use and cite the upstream repositories. This mirror only pins the revision our engines were built and measured against.

Original authors

incoai: DFlash2 draft model for MiniMax-M3.1-Flash
MiniMax: target model MiniMax-M3.1-Flash
Base model: MiniMaxAI/MiniMax-M3.1-Flash@54e4e69dfc452b1e94a426c24a4fdeeaa8e3ad5e

License

The upstream repo declares Apache-2.0 but ships no LICENSE file, so LICENSE is the standard Apache License 2.0 text from apache.org. It is a draft for MiniMax-M3.1-Flash, whose own use is governed by the MiniMax Community License (see the MiniMax mirror).

How our engines use it

Downloaded for drafter comparisons with MiniMax M3.1 Flash on the two-node engine; it is not integrated into the production engine. See the engine README.

Files

5 files, 4,248,080,758 bytes (4.25 GB). README.md is this card; the upstream model card is kept verbatim as README.upstream.md. Large-file sha256 values equal the upstream LFS hashes and were recomputed from our local copies before upload (any mismatch would have stopped the upload). SHA256SUMS lists every file.

File	Bytes	sha256
`LICENSE` (added by this mirror)	11,358	`cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30`
`README.upstream.md` (upstream `README.md`)	4,771	`8ec2de932fdd0e2d72a36ba17a0c593cfcba053609661a22fcb841047ae93282`
`assets/dflash2-figure.png`	286,889	`6d8dcc9a9472bddb644c881fe83050b5343d09bf5ffb0912e5598e55fe671e99`
`config.json`	1,252	`64c32db3c3bc5de64a048d949c5acd91352ae9b0f02ef636e3382c838f1d0815`
`model.safetensors`	4,247,776,488	`ad861f4ddd8f7204cda1f407837403e9095e1fb1f36d4c9f4188185261cc4fd1`

All files of the upstream revision are mirrored.

Downloads last month: 18

Safetensors

Model size

2B params

Tensor type

BF16

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for hyukudan/MiniMax-M3.1-Flash-DFlash2-mirror

Base model

MiniMaxAI/MiniMax-M3.1-Flash

Quantized

(32)

this model