GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
blackbird0831
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
blackbird0831
Model Tree
Qwen/Qwen3-4B
Input Modalities
Output Modalities
Supported Functionality
Scored by a deterministic grade-band check (primary) + an LLM judge, all models under the same harness.
| Metric | base | tuned 0.6B | tuned 4B |
|---|---|---|---|
| Robustness, held-out (0-2) | 0.33 | 1.08 | 1.69 |
| Robustness, adversarial (0-2) | 0.23 | 0.53 | 1.53 |
| Task quality, adversarial (0-2) | 0.23 | 0.33 | 1.57 |
| Violation rate, adversarial | 86.7% | 73.3% | 20.0% |
| Mechanical band-fail, adversarial | 33.3% | 3.3% | 0.0% |
| Golden set (25 greedy prompts) | - | 24/25 | 25/25 |
The 4B holds its grade-7-8 reading level under jailbreak-style pushback where the base model caves. Full table + error analysis: https://github.com/blackbird-alt/vocab-locked-writing-tutor
python
from peft import PeftModelfrom transformers import AutoModelForCausalLM, AutoTokenizerbase="Qwen/Qwen3-4B"tok=AutoTokenizer.from_pretrained(base)m=AutoModelForCausalLM.from_pretrained(base, device_map="auto")m=PeftModel.from_pretrained(m, "blackbird0831/qwen3-4b-writing-tutor")msgs=[{"role":"system","content":"You are Billy-Bob-Joe, a friendly writing and grammar tutor for a 7th-8th grade student."}, {"role":"user","content":"what is a comma splice?"}]ids=tok.apply_chat_template(msgs, add_generation_prompt=True, return_tensors="pt").to(m.device)print(tok.decode(m.generate(ids, max_new_tokens=200)[0][ids.shape[1]:], skip_special_tokens=True))Certifies spec adherence (band + correctness + escalation-resistance), not measured learning outcomes. Still trips ~20% of the hardest multi-part jailbreaks.
from peft import PeftModelfrom transformers import AutoModelForCausalLM, AutoTokenizerbase="Qwen/Qwen3-4B"tok=AutoTokenizer.from_pretrained(base)m=AutoModelForCausalLM.from_pretrained(base, device_map="auto")m=PeftModel.from_pretrained(m, "blackbird0831/qwen3-4b-writing-tutor")msgs=[{"role":"system","content":"You are Billy-Bob-Joe, a friendly writing and grammar tutor for a 7th-8th grade student."}, {"role":"user","content":"what is a comma splice?"}]ids=tok.apply_chat_template(msgs, add_generation_prompt=True, return_tensors="pt").to(m.device)print(tok.decode(m.generate(ids, max_new_tokens=200)[0][ids.shape[1]:], skip_special_tokens=True))