EvoSteer-9b by Ta Be: Open-Weight Model

Runs On

What it takes to serve EvoSteer-9b (9.7B 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	19.3 GB	23.2 GB	1x MI300X (192 GB) Vultr	$1.85	1x H100 $1.99 · 1x MI325X $2.00
8-bit	9.7 GB	11.6 GB	1x MI300X (192 GB) Vultr	$1.85	1x H100 $1.99 · 1x MI325X $2.00
4-bit	4.8 GB	5.8 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 18, 2026.

EvoSteer-9b on every accelerator the SAVRN Index prices, at every precision

Model Card

By Ta Be, published under apache-2.0, revision 73a71805af1e.

The trained EvoSteer orchestrator as a complete model: the trained LoRA adapter (rank 64, alpha 128 on qproj, kproj, vproj, oproj, outproj and inprojqkv) is merged into the Qwen3.5-9B weights. The merged layers are stored in float32. The LoRA update is about 0.06% of the weight norm, and rounding the merged layers to bfloat16 would discard most of it, so load the model in float32 to reproduce the trained orchestrator exactly; all other layers are the unchanged bfloat16 base weights. evosteerheads.safetensors holds the learned value and flow heads of the same checkpoint (heads. and the published value snapshot publishedvalue.).

Read Ta Be's full model card

The trained EvoSteer orchestrator as a complete model: the trained LoRA adapter (rank 64, alpha 128 on q_proj, k_proj, v_proj, o_proj, out_proj and in_proj_qkv) is merged into the Qwen3.5-9B weights.

The merged layers are stored in float32. The LoRA update is about 0.06% of the weight norm, and rounding the merged layers to bfloat16 would discard most of it, so load the model in float32 to reproduce the trained orchestrator exactly; all other layers are the unchanged bfloat16 base weights.

import torch
from transformers import AutoTokenizer, Qwen3_5ForConditionalGeneration

model = Qwen3_5ForConditionalGeneration.from_pretrained("beita6969/EvoSteer-9b", dtype=torch.float32)
tokenizer = AutoTokenizer.from_pretrained("beita6969/EvoSteer-9b")

evosteer_heads.safetensors holds the learned value and flow heads of the same checkpoint (heads.* and the published value snapshot published_value.*).

Configuration

Architecture: Qwen3_5ForConditionalGeneration
Context length (tokens): 262,144
Layers: 32
Hidden size: 4,096
Feed-forward size: 12,288
Attention heads: 16
Key/value heads: 4
Head dimension: 256
Vocabulary size: 248,320
Model type: qwen3_5

Identity and Version

Repository: beita6969/EvoSteer-9b
Publisher: Ta Be
Task: Image and text to text
Modality: Image and text
Library: transformers
Parameters: 9.7B parameters
Languages: Not stated by the source
Revision: 73a71805af1e6e15aade8f0584229596da8fa53c
First published: 2026-10-01
Last updated: 2026-10-02

Files and Weights

17 files, 22.7 GB in total. The weights are 5 files totalling 22.7 GB in safetensors.

Weights5 files · 22.7 GB

Configuration4 files · 83.1 KB

Tokenizer4 files · 22.9 MB

Documentation2 files · 12.4 KB

Other1 file · 7.8 KB

Repository1 file · 1.6 KB

Every file

File	Type	Size	SHA-256
evosteer_heads.safetensors	Weights	12.7 MB	d25e9c669f1f
model.safetensors-00001-of-00004.safetensors	Weights	5.3 GB	4cbb1529edde
model.safetensors-00002-of-00004.safetensors	Weights	5.3 GB	2a95a94c3e83
model.safetensors-00003-of-00004.safetensors	Weights	7.2 GB	e28bf0206aa9
model.safetensors-00004-of-00004.safetensors	Weights	4.9 GB	8bdd656244f9
config.json	Configuration	2.7 KB	—
model.safetensors.index.json	Configuration	79.7 KB	—
preprocessor_config.json	Configuration	390 B	—
video_preprocessor_config.json	Configuration	385 B	—
LICENSE	Documentation	11.3 KB	—
README.md	Documentation	1.1 KB	—
chat_template.jinja	Other	7.8 KB	—
.gitattributes	Repository	1.6 KB	—
merges.txt	Tokenizer	3.4 MB	—
tokenizer.json	Tokenizer	12.8 MB	5f9e4d4901a9
tokenizer_config.json	Tokenizer	16.7 KB	—
vocab.json	Tokenizer	6.7 MB	—

License and Download

License: apache-2.0
Access: Open weights, no gate
Download size: 22.7 GB

Download from Ta Be

Released by Ta Be through its official repository on Hugging Face. Read the license.

Built From

Derived from Qwen/Qwen3.5-9B

Memory Requirements

Precision	Weights in memory
As published	22.7 GB
16-bit	19.3 GB
8-bit	9.7 GB
4-bit	4.8 GB

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

Questions About EvoSteer-9b

How much GPU memory does EvoSteer-9b need?

About 23.2 GB at 16-bit and 5.8 GB at 4-bit: the weights (9.7B parameters) plus a working margin. A long context needs more.

What is the cheapest GPU to run EvoSteer-9b 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 EvoSteer-9b commercially?

Yes. EvoSteer-9b 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 EvoSteer-9b's context length?

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

Similar Models

Model · Image and text to text

Qwen3.5-9B

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 9.7B parameters 262,144 tokens transformers

View model

Model · Image and text to text

Qwen3.5-9B-AWQ

QuantTrio

This repo quantizes the model using data-free quantization technique. As of 2026-02-25, make sure your system has cuda12.8 installed. Then, create a fresh Python environment (e.g. python3.12 venv) and run: 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…

Open weights apache-2.0 9.7B parameters 262,144 tokens transformers

View model

Model · Image and text to text

q35-sr-step24

Rong-Xi Tan

This repository contains the completed training step25 HF export. The repository name retains step24 for compatibility with the requested upload destination. Optimizer and scheduler state are not included.

Open weights 9.7B parameters 262,144 tokens transformers

View model

Model · Image and text to text

qwen35-9b-harvey-v4-notes-ntp-100m-kl-0p01

Violet Xiang

Full Qwen3.5-9B model from the September 22, 2026 Harvey notes-only training runs. This revision is epoch 2, step 1550 of a two-epoch run, job 1016164. The final checkpoint is on main and final; epoch 1 is on epoch1. Each saved checkpoint is also available by its checkpoint- reference. - 100,000,076 note labels/epoch before causal shift; 99,993,882 scored labels/epoch after shift. - 223,681 notes, packed into 6,200 rows of 16,384 tokens with seed 731. - 199,987,764 scored supervised labels seen through this checkpoint. - No supervised trajectory examples. These are the notes-only controls, separate from notes + trajectory mixture models. - Eight GH200 GPUs, effective batch 8, learning rate…

Open weights apache-2.0 9.7B parameters 262,144 tokens transformers

View model

Model · Image and text to text

qwen35-9b-harvey-v4-notes-ntp-100m

Violet Xiang

Full Qwen3.5-9B model from the September 22, 2026 Harvey notes-only training runs. This revision is epoch 2, step 1550 of a two-epoch run, job 1016041. The final checkpoint is on main and final; epoch 1 is on epoch1. Each saved checkpoint is also available by its checkpoint- reference. - 100,000,076 note labels/epoch before causal shift; 99,993,882 scored labels/epoch after shift. - 223,681 notes, packed into 6,200 rows of 16,384 tokens with seed 731. - 199,987,764 scored supervised labels seen through this checkpoint. - No supervised trajectory examples. These are the notes-only controls, separate from notes + trajectory mixture models. - Eight GH200 GPUs, effective batch 8, learning rate…

Open weights apache-2.0 9.7B parameters 262,144 tokens transformers

View model

Model · Image and text to text

qwen35-9b-harvey-v4-notes-ntp-30m

Violet Xiang

Full Qwen3.5-9B model from the September 22, 2026 Harvey notes-only training runs. This revision is epoch 2, step 466 of a two-epoch run, job 1016040. The final checkpoint is on main and final; epoch 1 is on epoch1. Each saved checkpoint is also available by its checkpoint- reference. - 29,999,869 note labels/epoch before causal shift; 29,998,010 scored labels/epoch after shift. - 68,799 notes, packed into 1,864 rows of 16,384 tokens with seed 731. - 59,996,020 scored supervised labels seen through this checkpoint. - No supervised trajectory examples. These are the notes-only controls, separate from notes + trajectory mixture models. - Eight GH200 GPUs, effective batch 8, learning rate…

Open weights apache-2.0 9.7B parameters 262,144 tokens transformers

View model