Tiiny Pocket verification
Imported from this repo onto a Tiiny Pocket (TiinyOS 0.1.39) and tested on 2026-11-17:
- Result: pass, 5/5 smoke tests passed
- NPU share 29%, listed size 19.0 GB (the device lists the source size)
- At 3725 prompt tokens: TTFT 2.94 s, prefill 1269.2 tok/s, decode 17.6 tok/s
Full report: tiiny-verification.md (raw data in tiiny-verification.json).
Repackaged for Tiiny Pocket import. A copy of empero-ai/Qwen3.5-9B-Claude-Opus-4.6-Distill at a0a71df249b8, made by tiiny-forge so the Tiiny's Hugging Face import accepts it as a fine-tune of the store model Qwen/Qwen3.5-9B. 379 tensors were cast from F16 to BF16 (the Tiiny's converter takes BF16 only); nothing else in the weights changed. The tensors the source lacks (listed below) are the base's, as published. Changes:
- config.json: the base's (Qwen/Qwen3.5-9B), same tensors, its transformers schema
- added 333 tensors from Qwen/Qwen3.5-9B (333: model.visual.* (333), 0.91 GB) as model-graft-from-base.safetensors, byte for byte; the source is a text-only build
- cast 379 tensors from F16 to BF16 (round to nearest even), as the converter requires
- split model.safetensors into 4 shards of at most 5 GB (the converter maps each file whole); tensors byte for byte
- merges.txt: copied from Qwen/Qwen3.5-9B
- preprocessor_config.json: copied from Qwen/Qwen3.5-9B
- video_preprocessor_config.json: copied from Qwen/Qwen3.5-9B
- vocab.json: copied from Qwen/Qwen3.5-9B
- model card pipeline_tag: text-generation -> image-text-to-text
Qwen3.5-9B Opus 4.6 Reasoning Distill
A reasoning-focused fine-tune of Qwen/Qwen3.5-9B by Empero AI, trained to produce detailed chain-of-thought reasoning inside <think> tags before providing its final answer.
This model was distilled from Claude Opus 4.6 and Qwen3.5 reasoning traces, teaching the model to think step-by-step through problems before responding.
GGUF quantizations available at: empero-ai/Qwen3.5-9B-Claude-Opus-4.6-Distill-GGUF
How to Use
from transformers import AutoTokenizer, AutoModelForCausalLM model_name = "empero-ai/qwen3.5-9b-opus-4.6-distill"tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)model = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto", trust_remote_code=True) messages = [ {"role": "system", "content": "You are a deep reasoning AI. When given a problem, you think through it carefully and methodically inside <think> tags before providing your final answer."}, {"role": "user", "content": "What is the probability of rolling at least one six in four dice rolls?"}] text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)inputs = tokenizer(text, return_tensors="pt").to(model.device)outputs = model.generate(**inputs, max_new_tokens=2048, temperature=0.7)print(tokenizer.decode(outputs[0], skip_special_tokens=True))
The model outputs its reasoning process inside <think> tags, followed by a clear final answer:
<think>Let me work through this step by step. The probability of NOT rolling a six on a single die is 5/6.The probability of NOT rolling a six on any of four rolls is (5/6)^4. (5/6)^4 = 625/1296 ≈ 0.4823 So the probability of rolling at least one six is:1 - 625/1296 = 671/1296 ≈ 0.5177 That's just over 50% — slightly more likely than not.</think> The probability of rolling at least one six in four dice rolls is **671/1296**, or approximately **51.77%**. This is calculated by finding the complement: the chance of getting no sixes at all is (5/6)⁴ ≈ 48.23%, so the chance of at least one six is 1 − (5/6)⁴ ≈ 51.77%.
Training Details
Method
- Base model: Qwen/Qwen3.5-9B
- Technique: QLoRA 4-bit (rank 64, alpha 128) — Supervised Fine-Tuning
- Epochs: 3 (best checkpoint selected at epoch 2)
- Hardware: RTX 5090 (32GB VRAM)
- Attention: SDPA
- Optimizer: Paged AdamW 8-bit
- Learning rate: 1e-4 with cosine schedule
- Effective batch size: 8
- Max sequence length: 4096
- Training time: ~4.5 hours
Results
Table with columns: Metric, Epoch 1, Epoch 2 (best), Epoch 3| Metric | Epoch 1 | Epoch 2 (best) | Epoch 3 |
|---|
| Eval Loss | 0.5205 | 0.4809 | 0.4915 |
| Eval Token Accuracy | 0.8494 | 0.8615 | 0.8617 |
| Eval Entropy | 0.508 | 0.434 | 0.394 |
Datasets
About Empero AI
This model was developed by Empero AI. We build open-source AI tools and models focused on advancing reasoning capabilities in smaller, efficient language models.
License
This model inherits the Apache 2.0 license from Qwen3.5-9B.