Humanizer-12B-Draftfit-DFlash is an open-weight model from JayYun, released under Apache License 2.0. It has 727M parameters and a 262,144-token context. At 16-bit it needs about 1.7 GB of GPU memory, which fits on 1x MI300X from $1.85 an hour, at the lowest prices in the SAVRN Index.
A validation checkpoint for Draftfit — train DFlash-family drafts jointly, convert compatible families, and continue training existing drafts. This DFlash draft accelerates Humanizer; it requires the target model. Target / teacher: Humanizer.
Runs On
What it takes to serve Humanizer-12B-Draftfit-DFlash (727M 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 | 1.5 GB | 1.7 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
| 8-bit | 0.7 GB | 0.9 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
| 4-bit | 0.4 GB | 0.4 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 23, 2026.
Humanizer-12B-Draftfit-DFlash on every accelerator the SAVRN Index prices, at every precision
Model Card
By JayYun, published under apache-2.0, revision 2a6b1b7bc5e9.
A validation checkpoint for Draftfit — train DFlash-family drafts jointly, convert compatible families, and continue training existing drafts. This DFlash draft accelerates Humanizer; it requires the target model. Target / teacher: Humanizer. Initial draft weights: Gemma4 DFlash. This is a target-specific speculative draft, not a standalone Humanizer fine-tune. 4/8-bit target with FP32 floating computation · 4-bit drafts. Median throughput includes prefill and verification; acceptance is accepted/proposed draft tokens. Same 10 held-out inputs ×2 repeats per model, greedy, up to 64 output tokens, draft cap 3, lookup off. All 160 requests matched target-only tokens, including…
Read JayYun's full model card
Draftfit · Humanizer 12B DFlash
A validation checkpoint for Draftfit — train DFlash-family drafts jointly, convert compatible families, and continue training existing drafts.
This DFlash draft accelerates Humanizer; it requires the target model.
Target / teacher: Humanizer. Initial draft weights: Gemma4 DFlash. This is a target-specific speculative draft, not a standalone Humanizer fine-tune.
Performance
| Model | tok/s | Acceptance | vs. original |
|---|---|---|---|
| Target only | 21.71 | — | — |
| Original DFlash | 23.43 | 30.1% | reference |
| Draftfit DSpark | 33.49 | 59.9% | +42.9% |
| Draftfit DFlash | 32.62 | 59.9% | +39.2% |
| Draftfit DFlash2 | 31.23 | 55.6% | +33.3% |
By domain · tok/s
| Model | General prose | Blog | Report | Essay | |
|---|---|---|---|---|---|
| Target only | 21.52 | 21.51 | 21.44 | 22.00 | 21.79 |
| Original DFlash | 25.22 | 24.38 | 23.75 | 22.01 | 22.98 |
| Draftfit DSpark | 33.59 | 39.29 | 33.54 | 31.77 | 30.93 |
| Draftfit DFlash | 32.92 | 38.81 | 32.80 | 29.75 | 29.13 |
| Draftfit DFlash2 | 33.16 | 35.81 | 30.46 | 31.14 | 29.48 |
Apple M5 Pro · MLX · mixed 4/8-bit target with FP32 floating computation · 4-bit drafts. Median throughput includes prefill and verification; acceptance is accepted/proposed draft tokens. Same 10 held-out inputs ×2 repeats per model, greedy, up to 64 output tokens, draft cap 3, lookup off. All 160 requests matched target-only tokens, including previous-checkpoint comparisons.
Continued drafts improved median throughput by 2.6–4.9% over their previous checkpoints in this run; trial variability is substantial, so this is not a robust or long-form speedup claim. Domain acceptance, previous checkpoints and details.
How to use
MLX — tested. After runtime setup, pass prose to rewrite:
hf download jayyun98/Humanizer-12B-Draftfit-DFlash run_mlx.py --local-dir ./humanizer-draft
python ./humanizer-draft/run_mlx.py --text "The meeting has been moved to Friday." --check
vLLM / SGLang — unverified for this checkpoint. No CUDA serving or speed claim; use the tested MLX path.
Basic precision showed occasional token differences; the supplied runner uses the tested FP32 setting. Provenance · Apache-2.0 · DFlash MIT · Attribution.
Configuration
- Architecture
- DFlashDraftModel
- Context length (tokens)
- 262,144
- Layers
- 5
- Hidden size
- 3,840
- Feed-forward size
- 7,680
- Attention heads
- 32
- Key/value heads
- 8
- Head dimension
- 128
- Vocabulary size
- 262,144
- Sliding window (tokens)
- 2,048
- RoPE base
- 1,000,000
- Model type
- qwen3
Identity and Version
- Repository
- jayyun98/Humanizer-12B-Draftfit-DFlash
- Publisher
- JayYun
- Task
- Not stated by the source
- Modality
- Other
- Library
- Not stated by the source
- Parameters
- 727M parameters
- Languages
- en, zh
- Revision
- 2a6b1b7bc5e936c4e942b025a1b23b52ce3ce860
- First published
- 2026-10-14
- Last updated
- 2026-10-15
Files and Weights
14 files, 1.5 GB in total. The weights are 1 file totalling 1.5 GB in safetensors.
Every file
| File | Type | Size | SHA-256 |
|---|---|---|---|
| model.safetensors | Weights | 1.5 GB | 9620b02d4dff |
| config.json | Configuration | 1.2 KB | — |
| evaluation.json | Configuration | 12.5 KB | — |
| run_mlx.py | Configuration | 2.2 KB | — |
| LICENSE | Documentation | 11.4 KB | — |
| LICENSE-DFlash | Documentation | 1.1 KB | — |
| LICENSE-mlx-dspark | Documentation | 1.1 KB | — |
| NOTICE | Documentation | 494 B | — |
| README.md | Documentation | 2.8 KB | — |
| evaluation.md | Documentation | 2.7 KB | — |
| runtime-setup.md | Documentation | 1.9 KB | — |
| training.md | Documentation | 1.3 KB | — |
| mlx-runtime.patch | Other | 29.0 KB | — |
| .gitattributes | Repository | 1.5 KB | — |
License and Download
- License
- apache-2.0
- Access
- Open weights, no gate
- Download size
- 1.5 GB
Released by JayYun through its official repository on Hugging Face. Read the license.
Built From
- Derived from jialinyyzz/humanizer
Memory Requirements
| Precision | Weights in memory |
|---|---|
| As published | 1.5 GB |
| 16-bit | 1.5 GB |
| 8-bit | 0.7 GB |
| 4-bit | 0.4 GB |
Weights only, from the published parameter count; the key-value cache and runtime add to this.
Questions About Humanizer-12B-Draftfit-DFlash
How much GPU memory does Humanizer-12B-Draftfit-DFlash need?
About 1.7 GB at 16-bit and 0.4 GB at 4-bit: the weights (727M parameters) plus a working margin. A long context needs more.
What is the cheapest GPU to run Humanizer-12B-Draftfit-DFlash on?
At 16-bit, 1x MI300X from $1.85 an hour; at 4-bit, 1x MI300X from $1.85 an hour, at the lowest on-demand prices the SAVRN Index lists.
Can I use Humanizer-12B-Draftfit-DFlash commercially?
Yes. Humanizer-12B-Draftfit-DFlash is released under Apache License 2.0. The Apache License 2.0 is a permissive open-source license. It permits commercial use, modification and redistribution. It requires keeping the license and copyright notices and any NOTICE file, stating significant changes, and it includes an express patent grant from contributors.
What is Humanizer-12B-Draftfit-DFlash's context length?
262,144 tokens, from the maximum position embeddings in its published configuration.