grpo-qwen3.5-9b-seven-task is an open-weight model from Cyrus Kwan, released under Apache License 2.0. It has 9.4B parameters and a 262,144-token context. At 16-bit it needs about 22.6 GB of GPU memory, which fits on 1x MI300X from $1.85 an hour, at the lowest prices in the SAVRN Index.
Qwen/Qwen3.5-9B trained with grpo on the seven-task setting: 100 steps, 32 prompts x 8 rollouts per step, lr 2e-6, KL 1e-3, 32K rollout cap, verl v0.9.1 (FSDP2 + vLLM).
Runs On
What it takes to serve grpo-qwen3.5-9b-seven-task (9.4B 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 | 18.8 GB | 22.6 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
| 8-bit | 9.4 GB | 11.3 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
| 4-bit | 4.7 GB | 5.6 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 Oct 18, 2026.
grpo-qwen3.5-9b-seven-task on every accelerator the SAVRN Index prices, at every precision
Model Card
By Cyrus Kwan, published under apache-2.0, revision 9519eeffcac7.
Qwen/Qwen3.5-9B trained with grpo on the seven-task setting: 100 steps, 32 prompts x 8 rollouts per step, lr 2e-6, KL 1e-3, 32K rollout cap, verl v0.9.1 (FSDP2 + vLLM). Training data: 512 examples (seed 0) from the official training splits; system prompt "Solve the user's task and give the final answer directly."; reward = task correctness of the text after.
Read Cyrus Kwan's full model card
Qwen/Qwen3.5-9B trained with grpo on the seven-task setting: 100 steps, 32 prompts x 8 rollouts per step, lr 2e-6, KL 1e-3, 32K rollout cap, verl v0.9.1 (FSDP2 + vLLM). Training data: 512 examples (seed 0) from the official training splits; system prompt "Solve the user's task and give the final answer directly."; reward = task correctness of the text after </think>.
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
- wckwan/grpo-qwen3.5-9b-seven-task
- Publisher
- Cyrus Kwan
- Task
- Not stated by the source
- Modality
- Other
- Library
- Not stated by the source
- Parameters
- 9.4B parameters
- Languages
- rl
- Revision
- 9519eeffcac7c55239829e971d47dec263e8a81a
- First published
- 2026-09-25
- Last updated
- 2026-09-25
Files and Weights
9 files, 37.7 GB in total. The weights are 1 file totalling 37.6 GB in safetensors.
Every file
| File | Type | Size | SHA-256 |
|---|---|---|---|
| model.safetensors | Weights | 37.6 GB | 639cec2cfc0a |
| config.json | Configuration | 2.9 KB | — |
| generation_config.json | Configuration | 204 B | — |
| processor_config.json | Configuration | 1.2 KB | — |
| README.md | Documentation | 506 B | — |
| chat_template.jinja | Other | 7.8 KB | — |
| .gitattributes | Repository | 1.6 KB | — |
| tokenizer.json | Tokenizer | 20.0 MB | 06b9509352d2 |
| tokenizer_config.json | Tokenizer | 1.2 KB | — |
License and Download
- License
- apache-2.0
- Access
- Open weights, no gate
- Download size
- 37.6 GB
Released by Cyrus Kwan 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 | 37.6 GB |
| 16-bit | 18.8 GB |
| 8-bit | 9.4 GB |
| 4-bit | 4.7 GB |
Weights only, from the published parameter count; the key-value cache and runtime add to this.
Questions About grpo-qwen3.5-9b-seven-task
How much GPU memory does grpo-qwen3.5-9b-seven-task need?
About 22.6 GB at 16-bit and 5.6 GB at 4-bit: the weights (9.4B parameters) plus a working margin. A long context needs more.
What is the cheapest GPU to run grpo-qwen3.5-9b-seven-task 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 grpo-qwen3.5-9b-seven-task commercially?
Yes. grpo-qwen3.5-9b-seven-task 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 grpo-qwen3.5-9b-seven-task's context length?
262,144 tokens, from the maximum position embeddings in its published configuration.