GLM-5.2 is live. #1 throughput on OpenRouter, pay-per-token on FriendliAI. Try it today ➜
HamzaSaraswat
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
HamzaSaraswat
Model Tree
unsloth/Qwen3.6-27B
Input Modalities
Output Modalities
Supported Functionality
Rules are stable until at least 2029; salary figures change daily. So the rules go in the weights and the numbers stay in the prompt — you paste a current cap sheet and the model reasons over what it is handed. About an eighth of the training data uses invented future thresholds where answering from memory is wrong by construction, specifically to train the "read, don't recall" reflex.
499 held-out questions, graded programmatically against a deterministic CBA calculator (temperature 0, thinking disabled):
| Measure | Base Qwen3.6-27B | This adapter |
|---|---|---|
| Verdict accuracy (legal / illegal) | 58.0% | 78.0% |
| Arithmetic (every required figure exact) | 53.5% | 54.3% |
| Grounding (no invented figures) | 3.8% | 31.7% |
| Staleness probes (pasted figures must win) | 3.4% | 32.2% |
Strong on rule application — exception eligibility 97.8% verdicts / 100% arithmetic; buyout market, draft penalties, and exception surveys at or near 100%. Weak on long chained arithmetic (multi-bracket tax bills), the familiar limit of doing multi-step multiplication in-weights; the fix is tool use, not more training.
6,240 examples, none of them written freehand by a language model. A deterministic engine computes each answer and records every step; a narrator writes the prose; a verifier rejects any output containing a figure the engine never computed. Players are invented — attaching fabricated salaries to real people would teach false facts, which is the exact failure this design avoids.
python
from peft import PeftModelfrom transformers import AutoModelForCausalLM, AutoTokenizerbase = AutoModelForCausalLM.from_pretrained("unsloth/Qwen3.6-27B", dtype="bfloat16", device_map="auto")model = PeftModel.from_pretrained(base, "HamzaSaraswat/hardcap-qwen3.6-27b-lora")tokenizer = AutoTokenizer.from_pretrained("HamzaSaraswat/hardcap-qwen3.6-27b-lora")
Or serve it with vLLM, which hot-swaps adapters without reloading the base:
bash
vllm serve unsloth/Qwen3.6-27B --enable-lora \--lora-modules capologist=HamzaSaraswat/hardcap-qwen3.6-27b-lora --max-lora-rank 32
Use the system prompt from the repository (datagen/prompts.py) — the behavior described
there is what the model was trained against.
LoRA r=32, alpha=64, BF16, on q/k/v/o and gate/up/down projections. 2 epochs, lr 1e-4 cosine, effective batch 16, sequence length 8192, ~8.5h on one DGX Spark (GB10).
Loss is masked to response tokens only. An earlier run without that masking scored 26% on verdicts — worse than the base model — because most of the gradient went into reproducing an identical system prompt and predicting random cap-sheet salaries. That one flag was worth 52 points.