GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
andresnowak
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
andresnowak
Model Tree
Qwen/Qwen3.8-27B
Input Modalities
Output Modalities
Supported Functionality
| Base model | Qwen/Qwen3.8-27B |
| Algorithm | GRPO with the Dr.GRPO correction (advantages not divided by group std) |
| Framework | NeMo RL |
| Data | AceReason-Math |
| Steps | 100 |
| Learning rate | 1e-6, cosine decay to 1e-7, 10 warmup iters |
| Weight decay | 0.0 |
| Parallelism | TP=4, PP=2 |
text
reward = correctnessNo monitorability term. The treatment arm adds
× (1 + 0.1 × monitor_score), where the score comes from a weak monitor model
(HuggingFaceTB/SmolLM3-3B) reading the reasoning trace.
Batch shape is 16 prompts x 8 generations = 128 samples per step, over 100 steps.
| per sample | per step | total (100 steps) | |
|---|---|---|---|
| Total tokens (prompt + generated) | 4026 | ~515K | ~51.5M |
| Generated tokens only | 3873 | ~496K | ~49.6M |
Means are averaged over all 100 steps; prompts account for the ~153-token difference between the two rows.
python
from transformers import AutoModelForCausalLM, AutoTokenizerimport torch m = "andresnowak/qwen38-27b-math-grpo-baseline-drgrpo-step100"tok = AutoTokenizer.from_pretrained(m)model = AutoModelForCausalLM.from_pretrained(m, dtype=torch.bfloat16, device_map="auto")from transformers import AutoModelForCausalLM, AutoTokenizerimport torch m = "andresnowak/qwen38-27b-math-grpo-baseline-drgrpo-step100"tok = AutoTokenizer.from_pretrained(m)model = AutoModelForCausalLM.from_pretrained(m, dtype=torch.bfloat16, device_map="auto")