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
Open-weight model · Text generation
APUS-OpenJev-v1-35B-A3B
by APUS AI apus-ailab/APUS-OpenJev-v1-35B-A3B
APUS-OpenJev-v1-35B-A3B is an open-weight model for text generation from APUS AI, released under Apache License 2.0. It has 36B parameters and a 262,144-token context. At 16-bit it needs about 86.3 GB of GPU memory, which fits on 1x MI300X from $1.85 an hour, at the lowest prices in the SAVRN Index.
A Qwen3.5 MoE decision model for choosing browser actions, selecting workflow steps, and judging natural-language criteria.
Parameters36B
Context262,144
Weights71.9 GB
Licenseapache-2.0
AccessOpen weights
Monthly Downloads—
Runs On
What it takes to serve APUS-OpenJev-v1-35B-A3B (36B 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 | 71.9 GB | 86.3 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x MI325X $2.00 · 1x MI355X $2.59 |
| 8-bit | 36.0 GB | 43.1 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
| 4-bit | 18.0 GB | 21.6 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.
APUS-OpenJev-v1-35B-A3B on every accelerator the SAVRN Index prices, at every precision
Model Card
By APUS AI, published under apache-2.0, revision 494738529137.
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…
Read APUS AI's full model card
English | 中文 · Collection · Model family · Technical Report · Runtime
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.
Choose a decision budget
| Mode | Decoder layers | Intended use |
|---|---|---|
effort="high" |
40 | Primary release mode for decisions and text generation |
effort="low" |
20 | Experimental shallow decision exit, selected explicitly |
The decision interface accepts 2–16 request-specific candidates. Applications can use the returned candidate IDs to dispatch actions or build structured workflow results. See runtime instructions and validation status for the portable interface and its exact input contract.
Download and run
python -m pip install huggingface_hub
hf download apus-ailab/APUS-OpenJev-v1-35B-A3B --local-dir ./APUS-OpenJev-v1-35B-A3B
cd APUS-OpenJev-v1-35B-A3B
python -m pip install -r requirements.txt
python examples.py . --device cuda:0 --effort high
The examples use the portable reference runtime; its current acceptance status is recorded in RUNTIME.md. For reproducibility, pass an immutable repository commit with hf download --revision <commit>.
To access the underlying full-depth Transformers model using the validated loader:
from openjet_runtime import OpenJet
runtime = OpenJet.from_pretrained(".", device="cuda:0", dtype="bfloat16")
model = runtime.model
tokenizer = runtime.tokenizer
Use the pinned dependencies in requirements.txt. The bundled loader preserves sensitive checkpoint parameters in FP32 while loading the main weights in BF16. It also provides the 20-layer exit and candidate-distribution interface. A rough planning estimate is 75 GB or more GPU memory for BF16 inference, varying with prompt length, batching and runtime. This release's native evaluation used an RTX PRO 6000 with 96 GB; the estimate is not a tested minimum.
Evaluation and training
| This checkpoint's merged model | Correct / 80 | Accuracy |
|---|---|---|
| Native full 40 layers | 71 / 80 | 88.75% |
| Native experimental 20 layers | 63 / 80 | 78.75% |
The Frozen80 panel covers Browser, HelpSteer3, BoolQ, MNLI and attribute decisions. These results come from the merged checkpoint-5949 native GPU evaluation, not the source adapter's larger evaluation panels. Frozen80 is a reused engineering development panel, not a blind benchmark or an end-to-end browser success rate.
Training used 5,949 curriculum records from 3,898 parent groups, with LoRA on attention and shared experts while routed experts and the router remained frozen. See training details. Numerical merge differences, experimental-mode boundaries and validation status are collected in RUNTIME.md.
Series and license
The Collection groups the independent 4B, 9B and 35B-A3B repositories. Each model has its own checkpoint identity, evaluation and download counter. The family report provides the shared research context.
We thank the Qwen/Qwen3.5-35B-A3B team. Model licensing follows Apache-2.0; retain the accompanying license and provenance files when redistributing.
Authors: gumpcheng (xDAN2099), zhangxu, APUS AI-LAB.
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
Identity and Version
- Repository
- apus-ailab/APUS-OpenJev-v1-35B-A3B
- Publisher
- APUS AI
- Task
- Text generation
- Modality
- Text
- Library
- transformers
- Parameters
- 36B parameters
- Languages
- en, zh
- Revision
- 4947385291375f6f5149b3156ff2811c65ebf020
- First published
- 2026-09-22
- Last updated
- 2026-09-22
Files and Weights
53 files, 71.9 GB in total. The weights are 14 files totalling 71.9 GB in safetensors.
Weights14 files · 71.9 GB
Configuration25 files · 261.7 KB
Tokenizer4 files · 22.9 MB
Documentation7 files · 43.0 KB
Other2 files · 7.9 KB
Repository1 file · 1.6 KB
Every file
| File | Type | Size | SHA-256 |
|---|---|---|---|
| model.safetensors-00001-of-00014.safetensors | Weights | 5.4 GB | a2ca66629909 |
| model.safetensors-00002-of-00014.safetensors | Weights | 5.4 GB | e5f181cf5af9 |
| model.safetensors-00003-of-00014.safetensors | Weights | 5.4 GB | ea644b3cbf8e |
| model.safetensors-00004-of-00014.safetensors | Weights | 5.4 GB | 39dab0aebfe1 |
| model.safetensors-00005-of-00014.safetensors | Weights | 5.4 GB | 8b8b78f59a24 |
| model.safetensors-00006-of-00014.safetensors | Weights | 5.4 GB | 8c573270dc02 |
| model.safetensors-00007-of-00014.safetensors | Weights | 5.4 GB | 290529c22df7 |
| model.safetensors-00008-of-00014.safetensors | Weights | 5.4 GB | 7fa60cc5c6c2 |
| model.safetensors-00009-of-00014.safetensors | Weights | 5.3 GB | 9a6e23ba4435 |
| model.safetensors-00010-of-00014.safetensors | Weights | 5.4 GB | bf1b31f0f711 |
| model.safetensors-00011-of-00014.safetensors | Weights | 5.4 GB | 354a7fdaf8c6 |
| model.safetensors-00012-of-00014.safetensors | Weights | 5.4 GB | 7b840741312c |
| model.safetensors-00013-of-00014.safetensors | Weights | 5.4 GB | 665b58f61ebf |
| model.safetensors-00014-of-00014.safetensors | Weights | 2.2 GB | e118bb228fb7 |
| config.json | Configuration | 3.5 KB | — |
| depth_config.json | Configuration | 286 B | — |
| evaluation/adapter-merged-comparison.json | Configuration | 1.3 KB | — |
| evaluation/cpu-merge.json | Configuration | 301 B | — |
| evaluation/dtype-validation.json | Configuration | 4.6 KB | — |
| evaluation/merged-frozen80.json | Configuration | 749 B | — |
| evaluation/shard-verification.json | Configuration | 19.4 KB | — |
| evaluation/standalone-comparison.json | Configuration | 654 B | — |
| evaluation/standalone-frozen80.json | Configuration | 782 B | — |
| evaluation/subsets.json | Configuration | 694 B | — |
| evaluation/text-smoke.json | Configuration | 738 B | — |
| examples.py | Configuration | 2.3 KB | — |
| generation_config.json | Configuration | 244 B | — |
| merge-provenance.json | Configuration | 519 B | — |
| model.safetensors.index.json | Configuration | 187.5 KB | — |
| openjet_runtime/__init__.py | Configuration | 52 B | — |
| openjet_runtime/candidate_projection.py | Configuration | 4.4 KB | — |
| openjet_runtime/checkpoint_dtype.py | Configuration | 2.5 KB | — |
| openjet_runtime/contracts.py | Configuration | 3.8 KB | — |
| openjet_runtime/early_exit.py | Configuration | 8.0 KB | — |
| openjet_runtime/runtime.py | Configuration | 8.6 KB | — |
| preprocessor_config.json | Configuration | 390 B | — |
| release-manifest.json | Configuration | 9.0 KB | — |
| runtime-source-provenance.json | Configuration | 942 B | — |
| video_preprocessor_config.json | Configuration | 385 B | — |
| CODE-LICENSE.txt | Documentation | 11.4 KB | — |
| LICENSE | Documentation | 11.5 KB | — |
| NOTICE | Documentation | 551 B | — |
| README.md | Documentation | 4.6 KB | — |
| README.zh-CN.md | Documentation | 4.2 KB | — |
| RUNTIME.md | Documentation | 7.2 KB | — |
| training.md | Documentation | 3.7 KB | — |
| chat_template.jinja | Other | 7.8 KB | — |
| requirements.txt | Other | 176 B | — |
| .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
- 71.9 GB
Download from APUS AI
Released by APUS AI through its official repository on Hugging Face. Read the license.
Built From
- Derived from Qwen/Qwen3.5-35B-A3B
Memory Requirements
| Precision | Weights in memory |
|---|---|
| As published | 71.9 GB |
| 16-bit | 71.9 GB |
| 8-bit | 36.0 GB |
| 4-bit | 18.0 GB |
Weights only, from the published parameter count; the key-value cache and runtime add to this.
Questions About APUS-OpenJev-v1-35B-A3B
How much GPU memory does APUS-OpenJev-v1-35B-A3B need?
About 86.3 GB at 16-bit and 21.6 GB at 4-bit: the weights (36B parameters) plus a working margin. A long context needs more.
What is the cheapest GPU to run APUS-OpenJev-v1-35B-A3B 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 APUS-OpenJev-v1-35B-A3B commercially?
Yes. APUS-OpenJev-v1-35B-A3B 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 APUS-OpenJev-v1-35B-A3B's context length?
262,144 tokens, from the maximum position embeddings in its published configuration.
Similar Models
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…
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…
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…