Downloads · 30 days
15
21% of all-time downloads
Kimi-K2.6-NVFP4-W4A16-mixed
MPraveen/Kimi-K2.6-NVFP4-W4A16-mixed
Kimi-K2.6-NVFP4-W4A16-mixed is a machine learning model from MPraveen. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as other.
All-time downloads
73
Public
Repo size
595 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors595 GB · 100%
At a glance
- License
- other
- Model type
- kimi_k25
- Access
- Public
- Created
- Jun 7, 2026
- Updated
- Jun 8, 2026
- SHA
- 1508c9c5
Base models
Model card
From the Hugging Face model README
Kimi-K2.6 — NVFP4 W4A16 Quantized
Quantized with NVIDIA ModelOpt v0.44.0.
Source Model
| Field | Value |
|---|---|
| Base model | moonshotai/Kimi-K2.6 |
| Architecture | DeepSeekV3 MoE (671B total, ~37B active) |
| Layers | 61 (1 dense + 60 MoE) |
| Experts | 256 routed per layer, top-8 routing |
| Hidden dim | 7168 |
Quantization
| Field | Value |
|---|---|
| Method | Post-Training Quantization (PTQ) |
| Weight format | NVFP4 (4-bit NVIDIA FP4) |
| Activation format | BF16 (W4A16) |
| Expert weights | NVFP4 with FP8 per-block scale + FP32 per-tensor scale |
| Non-expert weights | BF16 (attention, dense, lm_head) |
| Tool | NVIDIA ModelOpt 0.44.0 |
Calibration
| Field | Value |
|---|---|
| Datasets | CNN/DailyMail, OpenCodeReasoning, OpenMathReasoning, Magpie-Pro, WikiText-103 |
| Samples | 128 + 96 + 96 + 96 + 96 = 512 |
| Sequence length | 512 tokens |
| Batch size | 1 |
Checkpoint
| Field | Value |
|---|---|
| Shards | 60 safetensors files |
| Total size | 595.2 GB |
| Quantization config | "nvfp4" |
Hardware
Quantized on 8× NVIDIA B200 (178 GB HBM3e each).
Generated: 2026-06-07 02:29 UTC
- Type
- kimi_k25
- License
- other
- Created
- Jun 7, 2026
- Updated
- Jun 8, 2026