HuggingFaceH4 / text-generation updated 1 year ago

zephyr-7b-beta

Name: zephyr-7b-beta
Author: HuggingFaceH4

Zephyr is a series of language models that are trained to act as helpful assistants. Zephyr-7B-β is the second model in the series, and is a fine-tuned version of mistralai/Mistral-7B-v0.1 that was trained on on a mix of publicly available, synthetic datasets using Direct Preference Optimization (DPO). We found that removing the in-built alignment of these datasets boosted performance on MT Bench and made the model m...

Params

7.2 B

Context

32,768

Downloads 30d

161 K

Likes

1,847

Commercial use: allowed mit Not gated SAFETENSORS 1 languages View on Hugging Face ↗

Download history

daily snapshots · 10 days

164 K161 K

Jul 28Jul 31Aug 3Aug 6

Can you run it?

Estimated VRAM at 8K context unless noted. Pick your hardware to see the verdict per quantization.

File	Quant	Size	Est. VRAM	Verdict on RTX 4090 · 24 GB
model.safetensors	bf16	14.5 GB	17.5 GB	✅ Runs comfortably
model.safetensors (bf16, full)	bf16 + 32K ctx	14.5 GB	20.8 GB	⚠️ Tight — reduce context

Estimate: file size × 1.1 + KV cache at 8K + 0.5 GB overhead. Not a benchmark — how we calculate this.

Run it

copy-paste, exact tags checked against the Hub

~ · curl · api/v1

$ curl -s https://aimodelscomparison.com/api/v1/models/zephyr-7b-beta
{
  "hf_id": "HuggingFaceH4/zephyr-7b-beta",
  "params_b": 7.24,
  "context_length": 32768,
  "license": { "id": "mit", "commercial": "yes" },
  "downloads_30d": 160703,
  "vram_estimates": [
    { "quant": "bf16", "gb": 17.5 }
  ],
  "updated_at": "2026-07-28T18:08:54Z"
}

est. VRAM —on RTX 4090 · 24 GBJSON API →

Specifications

Architecture: MistralForCausalLM
Parameters: 7.2 B
Tensor type: BF16
Context length: 32,768
Vocabulary: 32,000
Layers / heads: 32 / 32
Licence: mit
First seen on the Hub: 2023-10-26
Base model: Mistral-7B-v0.1
Training datasets: HuggingFaceH4/ultrachat_200k, HuggingFaceH4/ultrafeedback_binarized
MT-Bench (reported): 7.34
AlpacaEval (reported): 0.906
Drop (3-Shot) (reported): 9.6624370805369
MMLU (5-Shot) (reported): 61.07
GSM8k (5-shot) (reported): 12.736921910538
HellaSwag (10-Shot) (reported): 84.355706034654
TruthfulQA (0-shot) (reported): 57.449169427629
Winogrande (5-shot) (reported): 77.742699289661
AI2 Reasoning Challenge (25-Shot) (reported): 62.030716723549
Added to our catalog: 2026-07-28

Family

Base model and the most-downloaded derivatives in the catalog.

Base Mistral-7B-v0.1 mistralai 679 K

Compare with

Qwen3-8B 16.3 M Qwen2.5-7B-Instruct 12.1 M Qwen3.6-35B-A3B-NVFP4 11.3 M Llama-3.1-8B-Instruct 8.0 M Qwen2.5-7B-Instruct-AWQ 4.6 M granite-4.1-8b 4.2 M

Appears in

Best open coding models Commercial-use text models Models that fit in 12 GB