Uncensored version of Qwen/Qwen3.6-35B-A3B with refusal behavior removed via abliteration (norm-preserving orthogonalization). Zero refusals on harmful prompts. No false refusals on harmless prompts. Abliteration identifies the "refusal direction" in the model's residual stream — the linear direction that activates when the model decides to refuse — and surgically removes it from all output projection weights using norm-preserving orthogonalization. 1. Collect residual stream activations (last token position) for 512 harmful + 512 harmless prompts across all 40 layers 2. Compute mean difference vector per layer → this is the "refusal direction" candidate 3. Score layers by…
Cyber-F1-smoke is an open-weight model for text generation from Autumn. It has 35.1B parameters and a 262,144-token context. At 16-bit it needs about 84.3 GB of GPU memory, which fits on 1x MI300X from $1.85 an hour, at the lowest prices in the SAVRN Index. It draws 43 downloads a month.
This model is a fine-tuned version of DuyTa/Cyber-F1. It has been trained using TRL. This model was trained with SFT. - PEFT 0.21.0
Parameters35.1B
Context262,144
Weights24.4 GB
License—
AccessOpen weights
Monthly Downloads43
Runs On
What it takes to serve Cyber-F1-smoke (35.1B 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 | 70.2 GB | 84.3 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x MI325X $2.00 · 1x MI355X $2.59 |
| 8-bit | 35.1 GB | 42.1 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
| 4-bit | 17.6 GB | 21.1 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 Sep 24, 2026.
Cyber-F1-smoke on every accelerator the SAVRN Index prices, at every precision
Model Card
This model is a fine-tuned version of DuyTa/Cyber-F1. It has been trained using TRL. This model was trained with SFT. - PEFT 0.21.0
Excerpt from the card by Autumn.
Configuration
- Architecture
- Qwen3_5MoeForConditionalGeneration
- Context length (tokens)
- 262,144
- Layers
- 40
- Hidden size
- 2,048
- Attention heads
- 16
- Key/value heads
- 2
- Head dimension
- 256
- Vocabulary size
- 248,320
- Experts
- 256
- Experts active per token
- 8
- Model type
- qwen3_5_moe
- Quantization
- compressed-tensors
Identity and Version
- Repository
- autumn10/Cyber-F1-smoke
- Publisher
- Autumn
- Task
- Text generation
- Modality
- Text
- Library
- peft
- Parameters
- 35.1B parameters
- Languages
- sft, trl
- Revision
- 92025cd0fde4d28b43d20bf41c841113e13633f2
- First published
- 2026-09-17
- Last updated
- 2026-09-21
Files and Weights
13 files, 24.5 GB in total. The weights are 1 file totalling 24.4 GB in safetensors.
Weights1 file · 24.4 GB
Configuration3 files · 631.9 KB
Tokenizer2 files · 20.0 MB
Documentation1 file · 1.5 KB
Other5 files · 39.1 KB
Repository1 file · 1.6 KB
Every file
| File | Type | Size | SHA-256 |
|---|---|---|---|
| model.safetensors | Weights | 24.4 GB | d362b18b4d26 |
| config.json | Configuration | 602.9 KB | — |
| generation_config.json | Configuration | 142 B | — |
| recipe.yaml | Configuration | 28.9 KB | — |
| README.md | Documentation | 1.5 KB | — |
| chat_template.jinja | Other | 7.8 KB | — |
| runs/Sep17_19-09-35_3ec2326a4bfd/events.out.tfevents.1789672175.3ec2326a4bfd.17396.0 | Other | 7.7 KB | 4c8546ce896a |
| runs/Sep17_19-22-08_3ec2326a4bfd/events.out.tfevents.1789672928.3ec2326a4bfd.20038.0 | Other | 7.7 KB | 3cd57d8e9b91 |
| runs/Sep17_19-33-31_3ec2326a4bfd/events.out.tfevents.1789673611.3ec2326a4bfd.22253.0 | Other | 7.7 KB | 92a251bed757 |
| runs/Sep17_19-40-03_3ec2326a4bfd/events.out.tfevents.1789674003.3ec2326a4bfd.23764.0 | Other | 8.1 KB | fdb08c497e8e |
| .gitattributes | Repository | 1.6 KB | — |
| tokenizer.json | Tokenizer | 20.0 MB | 06b9509352d2 |
| tokenizer_config.json | Tokenizer | 1.1 KB | — |
License and Download
- License
- Not stated by the source
- Access
- Open weights, no gate
- Download size
- 24.4 GB
Download from Autumn
Released by Autumn through its official repository on Hugging Face.
Built From
- Adapter of DuyTa/Cyber-F1
- Derived from DuyTa/Cyber-F1
Memory Requirements
| Precision | Weights in memory |
|---|---|
| As published | 24.4 GB |
| 16-bit | 70.2 GB |
| 8-bit | 35.1 GB |
| 4-bit | 17.6 GB |
Weights only, from the published parameter count; the key-value cache and runtime add to this.
Questions About Cyber-F1-smoke
How much GPU memory does Cyber-F1-smoke need?
About 84.3 GB at 16-bit and 21.1 GB at 4-bit: the weights (35.1B parameters) plus a working margin. A long context needs more.
What is the cheapest GPU to run Cyber-F1-smoke 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.
What is Cyber-F1-smoke's context length?
262,144 tokens, from the maximum position embeddings in its published configuration.
Similar Models
Anansi-35B-A3B balances good instruct following, narrative reasoning, and sweet prose. Specifically for my 8gb potato, because I can't run the ~30b dense models, but maybe it can be good for your potato too? Anansi was created by combining the logical trengths of two great models via a 35/65 DARE-TIES merge, followed by a targeted (LM) head tensor interpolation to graft on better prose. I think I got the balance about right for what I was aiming for. Instruct following is much better, including with more complex instructions, most of the LM-head prose improvement comes through, and the reasoning is a little longer and still narrative/character-focused without becoming excessive. The A1…
Curated and trained by Eric Hartford, Lucas Atkins, and Fernando Fernandes, and Cognitive Computations This is our most spectacular outcome ever. FFT, all parameters, 16bit. 77.4 MMLU on 34b. And it talks like a dream. Although the max positional embeddings is 4k, we used rope theta of 1000000.0 and we trained with sequence length 8k. We plan to train on the upcoming 32k version as well. Our appreciation for the sponsors of Dolphin 2.9.1: - Crusoe Cloud - provided excellent on-demand 8xH100 node - OnDemand - provided inference sponsorship This model is based on Yi-1.5-34b, and is governed by apache 2.0 license. The base model has 4k context, but we used rope theta of 1000000.0 and the…
A Qwen3.5 MoE decision model for choosing browser actions, selecting workflow steps, and judging natural-language criteria. Give the model a shared state and a set of candidate actions; the included decision runtime returns a distribution over those candidates. This repository contains 35B-A3B checkpoint-5949 merged BF16 weights. It is a standalone model with root-level Hugging Face configuration and weights, requiring no separate LoRA adapter. The native merged model scores 71/80 (88.75%) at its full 40-layer depth on the Frozen80 development panel. The decision interface accepts 2–16 request-specific candidates. Applications can use the returned candidate IDs to dispatch actions or build…
Qwen2.5 is the latest series of Qwen large language models. For Qwen2.5, we release a number of base language models and instruction-tuned language models ranging from 0.5 to 72 billion parameters. Qwen2.5 brings the following improvements upon Qwen2: - Significantly more knowledge and has greatly improved capabilities in coding and mathematics, thanks to our specialized expert models in these domains. - Significant improvements in instruction following, generating long texts (over 8K tokens), understanding structured data (e.g, tables), and generating structured outputs especially JSON. More resilient to the diversity of system prompts, enhancing role-play implementation and…
Qwen2.5-Coder is the latest series of Code-Specific Qwen large language models (formerly known as CodeQwen). As of now, Qwen2.5-Coder has covered six mainstream model sizes, 0.5, 1.5, 3, 7, 14, 32 billion parameters, to meet the needs of different developers. Qwen2.5-Coder brings the following improvements upon CodeQwen1.5: - Significantly improvements in code generation, code reasoning and code fixing. Base on the strong Qwen2.5, we scale up the training tokens into 5.5 trillion including source code, text-code grounding, Synthetic data, etc. Qwen2.5-Coder-32B has become the current state-of-the-art open-source codeLLM, with its coding abilities matching those of GPT-4o. - A more…