---
base_model: StandardThinking/StandardOne-8B-SH
base_model_relation: quantized
library_name: transformers
license: apache-2.0
pipeline_tag: text-classification
tags:
  - mistral3
  - decision-model
  - schema-head
  - fp8
  - compressed-tensors
---

# Standard One 8B SH — FP8

**Version:** v4 (2026-10-13)

An FP8 (compressed-tensors, float8_e4m3 weights, dynamic per-token activations) quantization of the merged backbone
of [Standard One 8B SH](https://huggingface.co/StandardThinking/StandardOne-8B-SH) v4, with the same schema head.
The language-model linear projections (q/k/v/o, gate/up/down; 238 tensors) are quantized per output channel
(scale = absmax / 448); the vision tower, multi-modal projector, embeddings, norms and lm_head stay in BF16. This is
the same data-free recipe and `quantization_config` as
[StandardOne-8B-FP8](https://huggingface.co/StandardThinking/StandardOne-8B-FP8). The head (`head/`) is float32 and
identical to the BF16 release.

| | |
|---|---|
| Weights | ~10 GB (BF16 release: 17 GB) |
| Head | `head/schema_head.safetensors`, float32, taps: final layer and layer 25 |
| Prompt format | chat template with no default system message (see the [main card](https://huggingface.co/StandardThinking/StandardOne-8B-SH#prompt-format)) |

## Validation

Measured 2026-10-13 on one AMD Instinct MI350X with our SGLang-based decision engine (ROCm build, with the fix in the notes),
against the official Decision Index 0.3 run of the BF16 release (StandardOne-8B-SH v4) on the same requests. The sample takes 150
random requests from each of six benchmarks where the model is strong (Home appliance simulator, ContractNLI, HoVer, CLINC150, BANKING77, POP909-CL) and six where it is weak
(MuSR, WinoGrande, GSM8K, CRUXEval, ChessBench, MMLU-Pro); ContractNLI has 123 in total. Accuracy is the most probable answer against the benchmark's gold.

| Benchmarks | Questions | BF16 | FP8 (this repository) | Change (points) | Same answer as BF16 |
|---|---:|---:|---:|---:|---:|
| strong (6 benchmarks) | 4,209 | 93.09 % | 93.04 % | -0.05 | 99.45 % |
| weak (6 benchmarks) | 810 | 52.35 % | 52.72 % | +0.37 | 94.32 % |
| all | 5,019 | 86.51 % | 86.53 % | +0.02 | 98.63 % |

Answers change almost only where the model is unsure: on the strong benchmarks FP8 gives the same answer as BF16 for 99.5 % of questions, on the weak ones for 94.3 %, and accuracy stays within a fraction of a point in both. Per-benchmark numbers are in `release-manifest.json`.

## Notes

- Engines that load compressed-tensors FP8 (SGLang, vLLM) read this checkpoint. On ROCm with the native per-token
  activation-quantization fallback, SGLang 0.5.20 needs a small fix when the engine pads the token dimension
  (the per-token scale buffer is padded but the fallback writes unpadded rows); without it the server fails at
  warmup. The validation above used that fix.
- The PyTorch reference server in StandardOne-8B-SH is written for the BF16 checkpoint; use the BF16 release with it.

## License

Apache-2.0, the same terms as [StandardOne-8B-SH](https://huggingface.co/StandardThinking/StandardOne-8B-SH),
built on [Standard One 8B](https://huggingface.co/StandardThinking/StandardOne-8B) and
[Ministral 3 8B](https://huggingface.co/mistralai/Ministral-3-8B-Instruct-2512-BF16) (Apache-2.0).