Downloads · 30 days
123
81% of all-time downloads
minhhungg/qwen35-9b-p2s-merged-16bit
qwen35-9b-p2s-merged-16bit is a text generation model from minhhungg. Use it when you need the model to write or continue text. It is set up for transformers.
This repository contains the merged 16-bit checkpoint for the P2S-specialized Qwen3.5-9B model.
Downloads · 30 days
123
81% of all-time downloads
All-time downloads
152
Public
Parameters
9.7B
19.3 GB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors19.3 GB · 100%
How the weights are stored.
BF169.7B · 100%
From the Hugging Face model README
This repository contains the merged 16-bit checkpoint for the P2S-specialized Qwen3.5-9B model.
It is the high-precision merged Hugging Face artifact and the canonical source used for the downstream GGUF conversion workflow.
P2S Framework / SDK:
https://github.com/alberttrann/p2s-fuzzing-framework
| Resource | Link |
|---|---|
| P2S Framework / SDK | https://github.com/alberttrann/p2s-fuzzing-framework |
| P2S research artifacts / Golden-Silver records | https://huggingface.co/datasets/minhhungg/p2s-research-artifacts |
| AITasker specialization-data source | https://github.com/alberttrann/AITasker/tree/feat/Hung/SWT-Main |
| SEAL Track-A evaluation target | https://github.com/triet2809/SWP391_SealHackathon_BackEnd/tree/be-10/07/src/main/java/vn/edu/fpt/seal |
| LoRA adapter | https://huggingface.co/minhhungg/qwen35-9b-p2s-lora |
| Merged 4-bit checkpoint | https://huggingface.co/minhhungg/qwen35-9b-p2s-merged-4bit |
| Merged 16-bit checkpoint | https://huggingface.co/minhhungg/qwen35-9b-p2s-merged-16bit |
| GGUF F16 / Q8_0 | https://huggingface.co/minhhungg/p2s_gguf |
Qwen3.5-9B
+
P2S LoRA adapter
↓
THIS REPOSITORY
merged 16-bit checkpoint
│
└── llama.cpp conversion
↓
F16 GGUF
↓
Q8_0 GGUF
The model was specialized from P2S self-play records generated on AITasker.
| Corpus stage | Silver | Candidate-Golden | Total |
|---|---|---|---|
| Raw generator output | 1,917 | 44 unique | 1,961 |
| Deduplicated source corpus | 1,782 | 44 unique | 1,826 |
| Final SFT corpus | 1,782 | 484 training instances | 2,266 |
The 484 positive training instances are repetitions of the same 44 unique candidate-Golden records:
44 × 10 = 440
440 + one additional 44-record copy = 484
1,782 + 484 = 2,266
The 44 candidate-Goldens are training labels, not 44 independently verified vulnerabilities. A 5xx candidate represents server-fault evidence; a 2xx authorization/security-intent candidate requires stronger semantic verification before it can be called a verified security bypass.
| Setting | Value |
|---|---|
| Base | Qwen3.5-9B |
| Method | response-only SFT / LoRA |
| LoRA rank / alpha | 32 / 64 |
| LoRA dropout | 0.05 |
| Max sequence length | 24,576 |
| Batch / accumulation | 1 / 4 |
| Epochs | 6 |
| Peak LR | 2e-4 |
| Optimizer | AdamW 8-bit |
| Scheduler | cosine with restarts |
| Weight decay | 0.01 |
| NEFTune alpha | 5.0 |
| Precision | bfloat16 |
| Hardware | 1× A100 80 GB |
A real-batch mask check measured 92.6% prompt tokens masked from the supervised loss. The full training run used 3,402 optimization steps, adapted 58,195,968 / 9,468,009,712 parameters (0.61%), reported final loss 0.1418, peaked at 34.69 GB reserved GPU memory, and ran for approximately 8.3 hours.
Install a current Transformers/Accelerate stack:
pip install -U transformers accelerate
Then load this repository with the upstream-compatible Qwen3.5 model loader for your installed Transformers version.
For the most reproducible P2S workflow, follow:
https://github.com/alberttrann/p2s-fuzzing-framework
and the model/training reproduction document in docs/MODEL_AND_TRAINING.md.
The research workflow converts this merged checkpoint to F16 GGUF and then quantizes it to Q8_0 using llama.cpp.
Conceptually:
python convert_hf_to_gguf.py <MERGED_16BIT_DIR> --outfile qwen35-9b-p2s-f16.gguf --outtype f16
./llama-quantize qwen35-9b-p2s-f16.gguf qwen35-9b-p2s-Q8_0.gguf Q8_0
The resulting artifacts are published at:
https://huggingface.co/minhhungg/p2s_gguf
Use the 16-bit merged checkpoint when you want:
The reported Track-A evaluation uses the independent SEAL Spring Boot backend, not the AITasker source application.
| Metric | P2S FT | DeepSeek-V4-Flash-0731 | Base Qwen3.5-9B |
|---|---|---|---|
| Executed records | 1,075 | 1,094 | 1,122 |
| Candidate-Golden | 48 | 29 | 21 |
| Validated outcomes | 31 | 26 | 16 |
| ↳ validated 5xx server faults | 24 | 23 | 13 |
| ↳ verified security bypasses | 7 | 3 | 3 |
| Validated-outcome rate | 2.9% | 2.4% | 1.4% |
The reported P2S evaluation uses the Q8_0 GGUF artifact. Candidate-Golden records are not automatically vulnerabilities; Track A applies post-hoc semantic adjudication before security-bypass claims are made.
This is a research specialization for stateful REST API security testing, not a general-purpose assistant and not an autonomous vulnerability oracle.
Use it for:
Important limitations:
Do not use this artifact to test systems without authorization.
Tran Vo Minh Hung. P2S: Primitive-to-Semantics for Stateful REST API Security Testing Beyond Code Coverage. EAI FISAT 2026.