Qwen3.8-Flash-Next-RAZOR-65B-A6B-NVFP4 by Cleyesode

Runs On

What it takes to serve Qwen3.8-Flash-Next-RAZOR-65B-A6B-NVFP4 (88.1B parameters): the memory its weights need at each precision, and the cheapest way to rent enough data-center GPUs to hold them.

Precision	Weights	Memory needed	Cheapest setup	Per hour	Also fits
16-bit	176.2 GB	211.5 GB	1x MI325X (256 GB) Vultr	$2.00	1x MI355X $2.59 · 2x MI300X $3.70
8-bit	88.1 GB	105.7 GB	1x MI300X (192 GB) Vultr	$1.85	1x MI325X $2.00 · 1x MI355X $2.59
4-bit	44.1 GB	52.9 GB	1x MI300X (192 GB) Vultr	$1.85	1x H100 $1.99 · 1x MI325X $2.00

Memory is the weights at that precision plus 20% for the runtime and a short context; a long context needs more. Prices are the lowest on-demand hourly rates in the SAVRN Index, read Oct 12, 2026.

Qwen3.8-Flash-Next-RAZOR-65B-A6B-NVFP4 on every accelerator the SAVRN Index prices, at every precision

Model Card

Excellent 50% prune by Nickyang utilizing new RAZOR methodology. cleyesode further reduced memory requirements via nvfp4 quantization for use on Blackwell sm120 gpus. Quantization via NVIDIA/Model-Optimizer custom qwen4exp workflow: Please see post-quantization verificationreport for details. PLE isolated, full model repack for reduced memory pressure. Calibrated on curated samples from Nickyang/RazorCal: This model demonstrates very impressive retention of pre-prune / pre-quantize abilities - benchmarks in-progress. MTP is present but untested, updates will be posted as they become available. vLLM compatible via the latest branch: Example launch on 4 5060Ti 16GB GPUs (tensor parallel 4…

Excerpt from the card by Cleyesode, licensed other.

Configuration

Architecture: Qwen4ExpForConditionalGeneration
Context length (tokens): 262,144
Layers: 48
Hidden size: 2,560
Attention heads: 24
Key/value heads: 2
Head dimension: 256
Vocabulary size: 248,320
Experts: 256
Experts active per token: 10
Model type: qwen4_exp
Quantization: modelopt

Identity and Version

Repository: cleyesode/Qwen3.8-Flash-Next-RAZOR-65B-A6B-NVFP4
Publisher: Cleyesode
Task: Image and text to text
Modality: Image and text
Library: transformers
Parameters: 88.1B parameters
Languages: moe
Revision: 333f8a3d6b4d9f2d6565d45b6cc0bf34c812895b
First published: 2026-10-05
Last updated: 2026-10-05

Files and Weights

213 files, 98.7 GB in total. The weights are 193 files totalling 98.6 GB in safetensors.

Weights193 files · 98.6 GB

Configuration11 files · 16.8 MB

Tokenizer4 files · 22.9 MB

Documentation2 files · 16.0 KB

Other2 files · 31.2 KB

Repository1 file · 1.6 KB

Every file

File	Type	Size	SHA-256
model-00001-of-00065.safetensors	Weights	734.0 MB	9833e59b74a3
model-00002-of-00065.safetensors	Weights	734.0 MB	5f94efb0c3af
model-00003-of-00065.safetensors	Weights	734.0 MB	0cc2e723fe4e
model-00004-of-00065.safetensors	Weights	734.0 MB	a98b54b90993
model-00005-of-00065.safetensors	Weights	734.0 MB	5a685772ceb5
model-00006-of-00065.safetensors	Weights	734.0 MB	488658d77efa
model-00007-of-00065.safetensors	Weights	734.0 MB	cf33b2d9c16d
model-00008-of-00065.safetensors	Weights	734.0 MB	26f23ec2d362
model-00009-of-00065.safetensors	Weights	734.0 MB	35a264b8987e
model-00010-of-00065.safetensors	Weights	734.0 MB	061993f2654a
model-00011-of-00065.safetensors	Weights	734.0 MB	5890b85605b6
model-00012-of-00065.safetensors	Weights	734.0 MB	e3959640e6f7
model-00013-of-00065.safetensors	Weights	734.0 MB	73de529de7e3
model-00014-of-00065.safetensors	Weights	734.0 MB	19f6820c72c5
model-00015-of-00065.safetensors	Weights	734.0 MB	d8151fde6ae7
model-00016-of-00065.safetensors	Weights	734.0 MB	8e83144b428d
model-00017-of-00065.safetensors	Weights	734.0 MB	ffcd9ac13058
model-00018-of-00065.safetensors	Weights	734.0 MB	7141d4fb67e2
model-00019-of-00065.safetensors	Weights	734.0 MB	f686e6b00482
model-00020-of-00065.safetensors	Weights	734.0 MB	8dedd659f112
model-00021-of-00065.safetensors	Weights	734.0 MB	b7c7c947f551
model-00022-of-00065.safetensors	Weights	734.0 MB	44c043bb5f3f
model-00023-of-00065.safetensors	Weights	734.0 MB	981c39b2bc6e
model-00024-of-00065.safetensors	Weights	734.0 MB	dc2cdd8ecedc
model-00025-of-00065.safetensors	Weights	734.0 MB	301cbf7476a2
model-00026-of-00065.safetensors	Weights	734.0 MB	6f127ce6588e
model-00027-of-00065.safetensors	Weights	734.0 MB	603a5ffb62f5
model-00028-of-00065.safetensors	Weights	734.0 MB	ae5732402f2a
model-00029-of-00065.safetensors	Weights	734.0 MB	245e8167ab04
model-00030-of-00065.safetensors	Weights	734.0 MB	540f17bdbd68
model-00031-of-00065.safetensors	Weights	734.0 MB	4578248276f0
model-00032-of-00065.safetensors	Weights	734.0 MB	66e62e129895
model-00033-of-00065.safetensors	Weights	734.0 MB	74d54a44d43c
model-00034-of-00065.safetensors	Weights	734.0 MB	9ca93a9d503f
model-00035-of-00065.safetensors	Weights	734.0 MB	47a28583531c
model-00036-of-00065.safetensors	Weights	734.0 MB	35445907cfb9
model-00037-of-00065.safetensors	Weights	734.0 MB	0bd5fee03013
model-00038-of-00065.safetensors	Weights	734.0 MB	e05bb8220b81
model-00039-of-00065.safetensors	Weights	734.0 MB	ae5143577dc1
model-00040-of-00065.safetensors	Weights	734.0 MB	f4ef369f4a99
model-00041-of-00065.safetensors	Weights	734.0 MB	0fd84283a213
model-00042-of-00065.safetensors	Weights	734.0 MB	fe6390d7b962
model-00043-of-00065.safetensors	Weights	734.0 MB	76034450f2ee
model-00044-of-00065.safetensors	Weights	734.0 MB	17f5fbb890ae
model-00045-of-00065.safetensors	Weights	734.0 MB	db144b4fc396
model-00046-of-00065.safetensors	Weights	734.0 MB	125f2bfd53e5
model-00047-of-00065.safetensors	Weights	726.8 MB	550d13853ddd
model-00048-of-00065.safetensors	Weights	719.9 MB	51e23d13bf76
model-00049-of-00065.safetensors	Weights	693.5 MB	73064c34efd3
model-00050-of-00065.safetensors	Weights	730.8 MB	a13741cf261a
model-00051-of-00065.safetensors	Weights	446.1 MB	f5b7bd8c4f5d
model-00052-of-00065.safetensors	Weights	1.7 GB	a467f8747475
model-00053-of-00065.safetensors	Weights	153.5 MB	74b0043759b3
model-00054-of-00065.safetensors	Weights	838.9 MB	47525f98cadd
model-00055-of-00065.safetensors	Weights	724.3 MB	4c242eb9de67
model-00056-of-00065.safetensors	Weights	719.8 MB	be04a955513b
model-00057-of-00065.safetensors	Weights	729.1 MB	9f46d2d392cb
model-00058-of-00065.safetensors	Weights	730.7 MB	5653f4837f3d
model-00059-of-00065.safetensors	Weights	730.8 MB	9cc9200d6ddb
model-00060-of-00065.safetensors	Weights	724.2 MB	32c7ee3c1333
model-00061-of-00065.safetensors	Weights	711.0 MB	49d0c47c5201
model-00062-of-00065.safetensors	Weights	31.5 MB	cae8ca7562a2
model-00063-of-00065.safetensors	Weights	1.3 GB	c8af64f29a21
model-00064-of-00065.safetensors	Weights	24.4 MB	3354b0c77888
model-00065-of-00065.safetensors	Weights	1.3 GB	50be0ccd11e4
ple-embedding-fp8-00001-of-00128.safetensors	Weights	400.0 MB	eaca82dcfdd0
ple-embedding-fp8-00002-of-00128.safetensors	Weights	400.0 MB	47fc4902363b
ple-embedding-fp8-00003-of-00128.safetensors	Weights	400.0 MB	47c588435f37
ple-embedding-fp8-00004-of-00128.safetensors	Weights	400.0 MB	2c093738c6dd
ple-embedding-fp8-00005-of-00128.safetensors	Weights	400.0 MB	6557e77fdf57
ple-embedding-fp8-00006-of-00128.safetensors	Weights	400.0 MB	09069e924ea2
ple-embedding-fp8-00007-of-00128.safetensors	Weights	400.0 MB	39fdeb0e40a2
ple-embedding-fp8-00008-of-00128.safetensors	Weights	400.0 MB	2a5f8619bf71
ple-embedding-fp8-00009-of-00128.safetensors	Weights	400.0 MB	b798c7620cb2
ple-embedding-fp8-00010-of-00128.safetensors	Weights	400.0 MB	6ea7d498aac9
ple-embedding-fp8-00011-of-00128.safetensors	Weights	400.0 MB	1447e1ab9ec8
ple-embedding-fp8-00012-of-00128.safetensors	Weights	400.0 MB	9997a4a74bbe
ple-embedding-fp8-00013-of-00128.safetensors	Weights	400.0 MB	50e222b6c752
ple-embedding-fp8-00014-of-00128.safetensors	Weights	400.0 MB	93887729fc2f
ple-embedding-fp8-00015-of-00128.safetensors	Weights	400.0 MB	0c5ffa4343a9
ple-embedding-fp8-00016-of-00128.safetensors	Weights	400.0 MB	ae4ad241228b
ple-embedding-fp8-00017-of-00128.safetensors	Weights	400.0 MB	3932cefcae63
ple-embedding-fp8-00018-of-00128.safetensors	Weights	400.0 MB	5043680263ed
ple-embedding-fp8-00019-of-00128.safetensors	Weights	400.0 MB	60e5b1afefb0
ple-embedding-fp8-00020-of-00128.safetensors	Weights	400.0 MB	6517a67c87cb
ple-embedding-fp8-00021-of-00128.safetensors	Weights	400.0 MB	78ce36286524
ple-embedding-fp8-00022-of-00128.safetensors	Weights	400.0 MB	49cfc9e1dcc1
ple-embedding-fp8-00023-of-00128.safetensors	Weights	400.0 MB	b59e69dd4674
ple-embedding-fp8-00024-of-00128.safetensors	Weights	400.0 MB	2b6cb53fd59b
ple-embedding-fp8-00025-of-00128.safetensors	Weights	400.0 MB	ae5bd3657a85
ple-embedding-fp8-00026-of-00128.safetensors	Weights	400.0 MB	86291aee633b
ple-embedding-fp8-00027-of-00128.safetensors	Weights	400.0 MB	f989c6eb897d
ple-embedding-fp8-00028-of-00128.safetensors	Weights	400.0 MB	4861cf79e75b
ple-embedding-fp8-00029-of-00128.safetensors	Weights	400.0 MB	0836f7ed5db0
ple-embedding-fp8-00030-of-00128.safetensors	Weights	400.0 MB	9c5199903bcb
ple-embedding-fp8-00031-of-00128.safetensors	Weights	400.0 MB	8ae065e8be11
ple-embedding-fp8-00032-of-00128.safetensors	Weights	400.0 MB	a71eb18d5b05
ple-embedding-fp8-00033-of-00128.safetensors	Weights	400.0 MB	12b33b633415
ple-embedding-fp8-00034-of-00128.safetensors	Weights	400.0 MB	3156576996dd
ple-embedding-fp8-00035-of-00128.safetensors	Weights	400.0 MB	2086c93f9e5d
ple-embedding-fp8-00036-of-00128.safetensors	Weights	400.0 MB	cd927db32a7a
ple-embedding-fp8-00037-of-00128.safetensors	Weights	400.0 MB	0499963ba771
ple-embedding-fp8-00038-of-00128.safetensors	Weights	400.0 MB	2639eecb3169
ple-embedding-fp8-00039-of-00128.safetensors	Weights	400.0 MB	746986b61e27
ple-embedding-fp8-00040-of-00128.safetensors	Weights	400.0 MB	4285f5f7db31
ple-embedding-fp8-00041-of-00128.safetensors	Weights	400.0 MB	f0a9ca93b70f
ple-embedding-fp8-00042-of-00128.safetensors	Weights	400.0 MB	9aa83d2a28af
ple-embedding-fp8-00043-of-00128.safetensors	Weights	400.0 MB	461ec7cd9507
ple-embedding-fp8-00044-of-00128.safetensors	Weights	400.0 MB	6d9f4500e677
ple-embedding-fp8-00045-of-00128.safetensors	Weights	400.0 MB	1589b62bd85e
ple-embedding-fp8-00046-of-00128.safetensors	Weights	400.0 MB	c825681c528a
ple-embedding-fp8-00047-of-00128.safetensors	Weights	400.0 MB	b937f38a6c26
ple-embedding-fp8-00048-of-00128.safetensors	Weights	400.0 MB	1501aa5ed00f
ple-embedding-fp8-00049-of-00128.safetensors	Weights	400.0 MB	0adeab9489b6
ple-embedding-fp8-00050-of-00128.safetensors	Weights	400.0 MB	64f892eea4c5
ple-embedding-fp8-00051-of-00128.safetensors	Weights	400.0 MB	717b83d4dc3b
ple-embedding-fp8-00052-of-00128.safetensors	Weights	400.0 MB	13db03636f8b
ple-embedding-fp8-00053-of-00128.safetensors	Weights	400.0 MB	e49400562422
ple-embedding-fp8-00054-of-00128.safetensors	Weights	400.0 MB	988016a2b378
ple-embedding-fp8-00055-of-00128.safetensors	Weights	400.0 MB	2049412f66a3
ple-embedding-fp8-00056-of-00128.safetensors	Weights	400.0 MB	cfe7b4a5f206
ple-embedding-fp8-00057-of-00128.safetensors	Weights	400.0 MB	62f638adc06c
ple-embedding-fp8-00058-of-00128.safetensors	Weights	400.0 MB	b380b326d0c2
ple-embedding-fp8-00059-of-00128.safetensors	Weights	400.0 MB	0b871b0083f8
ple-embedding-fp8-00060-of-00128.safetensors	Weights	400.0 MB	ac4d6b53ac5c
ple-embedding-fp8-00061-of-00128.safetensors	Weights	400.0 MB	a808cd27213c
ple-embedding-fp8-00062-of-00128.safetensors	Weights	400.0 MB	6b8d7683d87c
ple-embedding-fp8-00063-of-00128.safetensors	Weights	400.0 MB	d3c67d74e41a
ple-embedding-fp8-00064-of-00128.safetensors	Weights	400.0 MB	4de5a666ca73
ple-embedding-fp8-00065-of-00128.safetensors	Weights	400.0 MB	452d010e8565
ple-embedding-fp8-00066-of-00128.safetensors	Weights	400.0 MB	44e7983afeaf
ple-embedding-fp8-00067-of-00128.safetensors	Weights	400.0 MB	90d3afa0b0c5
ple-embedding-fp8-00068-of-00128.safetensors	Weights	400.0 MB	3cbf695ece88
ple-embedding-fp8-00069-of-00128.safetensors	Weights	400.0 MB	b2a2df66940f
ple-embedding-fp8-00070-of-00128.safetensors	Weights	400.0 MB	5095baf4cc3e
ple-embedding-fp8-00071-of-00128.safetensors	Weights	400.0 MB	f75d0721dbf5
ple-embedding-fp8-00072-of-00128.safetensors	Weights	400.0 MB	67aff0eec9ce
ple-embedding-fp8-00073-of-00128.safetensors	Weights	400.0 MB	bed59e39c21a
ple-embedding-fp8-00074-of-00128.safetensors	Weights	400.0 MB	61ec59a17798
ple-embedding-fp8-00075-of-00128.safetensors	Weights	400.0 MB	5f45a1eb1642
ple-embedding-fp8-00076-of-00128.safetensors	Weights	400.0 MB	bfd6f01a07b8
ple-embedding-fp8-00077-of-00128.safetensors	Weights	400.0 MB	9c1512f0c71d
ple-embedding-fp8-00078-of-00128.safetensors	Weights	400.0 MB	72124234cb99
ple-embedding-fp8-00079-of-00128.safetensors	Weights	400.0 MB	201b80fc3d03
ple-embedding-fp8-00080-of-00128.safetensors	Weights	400.0 MB	f8da110caae9
ple-embedding-fp8-00081-of-00128.safetensors	Weights	400.0 MB	c2836137f832
ple-embedding-fp8-00082-of-00128.safetensors	Weights	400.0 MB	40151c138b0a
ple-embedding-fp8-00083-of-00128.safetensors	Weights	400.0 MB	0fd87a1dac4e
ple-embedding-fp8-00084-of-00128.safetensors	Weights	400.0 MB	ccb7defee3ce
ple-embedding-fp8-00085-of-00128.safetensors	Weights	400.0 MB	9bd50817bebf
ple-embedding-fp8-00086-of-00128.safetensors	Weights	400.0 MB	ccf7092bac50
ple-embedding-fp8-00087-of-00128.safetensors	Weights	400.0 MB	04d1fc292f74
ple-embedding-fp8-00088-of-00128.safetensors	Weights	400.0 MB	793da70b9e24
ple-embedding-fp8-00089-of-00128.safetensors	Weights	400.0 MB	e9279fbf865d
ple-embedding-fp8-00090-of-00128.safetensors	Weights	400.0 MB	a69fa55f346f
ple-embedding-fp8-00091-of-00128.safetensors	Weights	400.0 MB	68d002e926cd
ple-embedding-fp8-00092-of-00128.safetensors	Weights	400.0 MB	647a55a2df0e
ple-embedding-fp8-00093-of-00128.safetensors	Weights	400.0 MB	a71e32f7d712
ple-embedding-fp8-00094-of-00128.safetensors	Weights	400.0 MB	c851de18a0c3
ple-embedding-fp8-00095-of-00128.safetensors	Weights	400.0 MB	1ad487a9fced
ple-embedding-fp8-00096-of-00128.safetensors	Weights	400.0 MB	8634fe38f3d5
ple-embedding-fp8-00097-of-00128.safetensors	Weights	400.0 MB	82b8b6a65759
ple-embedding-fp8-00098-of-00128.safetensors	Weights	400.0 MB	7223f76f3f68
ple-embedding-fp8-00099-of-00128.safetensors	Weights	400.0 MB	29e64bddd54a
ple-embedding-fp8-00100-of-00128.safetensors	Weights	400.0 MB	ae01622d5163
ple-embedding-fp8-00101-of-00128.safetensors	Weights	400.0 MB	12b3902c1be1
ple-embedding-fp8-00102-of-00128.safetensors	Weights	400.0 MB	ae52847d84d1
ple-embedding-fp8-00103-of-00128.safetensors	Weights	400.0 MB	42f5465f6a91
ple-embedding-fp8-00104-of-00128.safetensors	Weights	400.0 MB	db2d6d1bac80
ple-embedding-fp8-00105-of-00128.safetensors	Weights	400.0 MB	c06f39ded422
ple-embedding-fp8-00106-of-00128.safetensors	Weights	400.0 MB	119fd985a9be
ple-embedding-fp8-00107-of-00128.safetensors	Weights	400.0 MB	908652f6ff8d
ple-embedding-fp8-00108-of-00128.safetensors	Weights	400.0 MB	9600553450d7
ple-embedding-fp8-00109-of-00128.safetensors	Weights	400.0 MB	34ac38e30747
ple-embedding-fp8-00110-of-00128.safetensors	Weights	400.0 MB	659c4b217a5b
ple-embedding-fp8-00111-of-00128.safetensors	Weights	400.0 MB	b54ace679dff
ple-embedding-fp8-00112-of-00128.safetensors	Weights	400.0 MB	7791352f911d
ple-embedding-fp8-00113-of-00128.safetensors	Weights	400.0 MB	4dadc9a59c83
ple-embedding-fp8-00114-of-00128.safetensors	Weights	400.0 MB	64cc8cca5502
ple-embedding-fp8-00115-of-00128.safetensors	Weights	400.0 MB	07e647add76a
ple-embedding-fp8-00116-of-00128.safetensors	Weights	400.0 MB	3edaf3c5a023
ple-embedding-fp8-00117-of-00128.safetensors	Weights	400.0 MB	e246fe57a94c
ple-embedding-fp8-00118-of-00128.safetensors	Weights	400.0 MB	8752e47ae40a
ple-embedding-fp8-00119-of-00128.safetensors	Weights	400.0 MB	42396ee03c7b
ple-embedding-fp8-00120-of-00128.safetensors	Weights	400.0 MB	144333fdf225
ple-embedding-fp8-00121-of-00128.safetensors	Weights	400.0 MB	d78e81074f5a
ple-embedding-fp8-00122-of-00128.safetensors	Weights	400.0 MB	5a843bb65cd1
ple-embedding-fp8-00123-of-00128.safetensors	Weights	400.0 MB	c8afbd9f9f93
ple-embedding-fp8-00124-of-00128.safetensors	Weights	400.0 MB	8f0cafc43499
ple-embedding-fp8-00125-of-00128.safetensors	Weights	400.0 MB	3d04031feb9c
ple-embedding-fp8-00126-of-00128.safetensors	Weights	400.0 MB	699e6e691bf7
ple-embedding-fp8-00127-of-00128.safetensors	Weights	400.0 MB	dcdf5a5a6a5c
ple-embedding-fp8-00128-of-00128.safetensors	Weights	400.0 MB	f4fed0d5f7b9
calibration_info.json	Configuration	609 B	—
config.json	Configuration	30.2 KB	—
generation_config.json	Configuration	178 B	—
hf_quant_config.json	Configuration	22.9 KB	—
kept_expert_indices.json	Configuration	111.0 KB	—
model.safetensors.index.json	Configuration	16.4 MB	dfcf19d2f83c
preprocessor_config.json	Configuration	337 B	—
pruning_info.json	Configuration	136.5 KB	—
release_info.json	Configuration	2.5 KB	—
verification_report.json	Configuration	2.9 KB	—
video_preprocessor_config.json	Configuration	332 B	—
LICENSE	Documentation	3.2 KB	—
README.md	Documentation	12.8 KB	—
chat_template.jinja	Other	9.0 KB	—
verification_report.txt	Other	22.2 KB	—
.gitattributes	Repository	1.6 KB	—
merges.txt	Tokenizer	3.4 MB	—
tokenizer.json	Tokenizer	12.8 MB	0997f410c57a
tokenizer_config.json	Tokenizer	16.4 KB	—
vocab.json	Tokenizer	6.7 MB	—

License and Download

License: other
Access: Open weights, no gate
Download size: 98.6 GB

Download from Cleyesode

Released by Cleyesode through its official repository on Hugging Face.

Built From

Derived from Nickyang/Qwen3.8-Flash-Next-RAZOR-65B-A6B-E256of512
Described by arXiv:2609.30465
Quantized from Nickyang/Qwen3.8-Flash-Next-RAZOR-65B-A6B-E256of512

Memory Requirements

Precision	Weights in memory
As published	98.6 GB
16-bit	176.2 GB
8-bit	88.1 GB
4-bit	44.1 GB

Weights only, from the published parameter count; the key-value cache and runtime add to this.

Questions About Qwen3.8-Flash-Next-RAZOR-65B-A6B-NVFP4

How much GPU memory does Qwen3.8-Flash-Next-RAZOR-65B-A6B-NVFP4 need?

About 211.5 GB at 16-bit and 52.9 GB at 4-bit: the weights (88.1B parameters) plus a working margin. A long context needs more.

What is the cheapest GPU to run Qwen3.8-Flash-Next-RAZOR-65B-A6B-NVFP4 on?

At 16-bit, 1x MI325X from $2.00 an hour; at 4-bit, 1x MI300X from $1.85 an hour, at the lowest on-demand prices the SAVRN Index lists.

What license is Qwen3.8-Flash-Next-RAZOR-65B-A6B-NVFP4 released under?

other, as its publisher declares it. Read the license text before commercial use.

What is Qwen3.8-Flash-Next-RAZOR-65B-A6B-NVFP4's context length?

262,144 tokens, from the maximum position embeddings in its published configuration.

Similar Models

Model · Image and text to text

DeepSeek-V4.1-Flash-oQ2.5e-mtp

Jun Kim

This model was quantized using oQ (oMLX v0.7.1) mixed-precision quantization with Improved oQe Quantization (jundot/omlx#4385). A mixed-precision MLX checkpoint of deepseek-ai/DeepSeek-V4.1-Flash, a multimodal mixture-of-experts model with 552B backbone parameters, a Causal Encoder-Decoder layout, Compressed Sparse Attention 2 (CSA2), Engram conditional memory and DSpark speculative decoding. It includes the language model, the Engram tables, the vision encoder, the multimodal aligner and the DSpark MTP weights. The effective average is 3.104 bits/weight excluding the Engram tables, or 3.592 bits/weight for the complete checkpoint. oQ2.5e allocates bits by measured layer sensitivity, then…

Open weights mit 102.4B parameters 1,048,576 tokens mlx

View model

Model · Image and text to text

GLM5.3-Flash-E192-DGX-Spark

AutoTrust AI Lab

autotrust/GLM5.3-Flash-E192-DGX-Spark is a compact build of zai-org/GLM-5.3-Flash for desktop Blackwell systems like NVIDIA DGX Spark. It's an unofficial derivative. It keeps 192 of the 288 routed experts in each layer by Neural Architecture Search (NAS), uses NVFP4 for the experts and still activates 18 B parameters per token. The weights take 124 GiB, the smallest in the family: two DGX Sparks connected by ConnectX-7 get about 62 GiB of weights per node and the most KV room of any build, and a single 180 GB Blackwell GPU (B200/GB200) runs it with 32 GiB of KV cache. The KV cache uses FP8. The repo includes a ready-to-use NVFP4 MTP draft in mtp-nvfp4/ (6.95 GB) for speculative decoding. On…

Open weights mit 110.7B parameters 1,048,576 tokens vllm

View model

Model · Image and text to text

tinfield-1-nvfp4-probe

Natapong Nitarach (Schwyter)

Public research conversion of badtheorylabs/Tinfield-1, derived from Qwen/Qwen3.8-Flash-Next. The source checkpoint was pinned to f7c714a97b71342db9e2fc04009d2704c656c80c. This Hugging Face repository contains the complete original export, including all 10 safetensors shards: - PLE embeddings, vision, MTP and other non-expert weights: retained in BF16. - The original configuration, tokenizer, processors and tensor index accompany the shards. The ple-int8-e579 Kaggle variant is a separate quantized PLE derivative. This Hugging Face checkpoint contains BF16 PLE. Load all shards and the configuration from this repository together. Conversion used NVIDIA Model-Optimizer recipe commit…

Open weights other 117B parameters 262,144 tokens transformers

View model

Model · Image and text to text

Qwen3.5-122B-A10B-FP8

Qwen

Over recent months, we have intensified our focus on developing foundation models that deliver exceptional utility and performance. Qwen3.5 represents a significant leap forward, integrating breakthroughs in multimodal learning, architectural efficiency, reinforcement learning scale, and global accessibility to empower developers and enterprises with unprecedented capability and efficiency. For more details, please refer to our blog post Qwen3.5. WMT24++: a harder subset of WMT24 after difficulty labeling and rebalancing; we report the averaged scores on 55 languages using XCOMET-XXL. Empty cells (--) indicate scores not yet available or not applicable. Empty cells (--) indicate scores not…

Open weights apache-2.0 125.1B parameters 262,144 tokens transformers

View model

Model · Image and text to text

GLM5.3-Flash-E224-DGX-Spark

AutoTrust AI Lab

autotrust/GLM5.3-Flash-E224-DGX-Spark is a compact build of zai-org/GLM-5.3-Flash for desktop Blackwell systems like NVIDIA DGX Spark. It's an unofficial derivative. It keeps 224 of the 288 routed experts in each layer by Neural Architecture Search (NAS), uses NVFP4 for the experts and still activates 18 B parameters per token. The weights take 141 GiB, which is small enough for two DGX Sparks connected by ConnectX-7 (256 GB of unified memory in total) with room left for long-context KV cache. A single 180 GB Blackwell GPU (B200/GB200) can also run it. The original GLM-5.3-Flash MTP layer ships unmodified in mtp/ (BF16) as an optional speculative-decoding draft. It gives about 1.85×…

Open weights mit 127.6B parameters 1,048,576 tokens vllm

View model

Model · Image and text to text

Qwen3.6-35B-A3B-FP8

Qwen

Following the February release of the Qwen3.5 series, we're pleased to share the first open-weight variant of Qwen3.6. Built on direct feedback from the community, Qwen3.6 prioritizes stability and real-world utility, offering developers a more intuitive, responsive, and genuinely productive coding experience. This release delivers substantial upgrades, particularly in For more details, please refer to our blog post Qwen3.6-35B-A3B. Empty cells (--) indicate scores not available or not applicable. For streamlined integration, we recommend using Qwen3.6 via APIs. Below is a guide to use Qwen3.6 via OpenAI-compatible API. Qwen3.6 can be served via APIs with popular inference frameworks. In…

Open weights apache-2.0 36B parameters 262,144 tokens transformers

View model