GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
sach0312
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Run this model inference with full control and performance in your environment.
Model Details
Model Provider
sach0312
Model Tree
Qwen/Qwen3-0.6B
Input Modalities
Output Modalities
Supported Functionality
| policy | avg P&L per game ($1 stakes) | games profitable | trades |
|---|---|---|---|
| model (greedy) | −0.0060 | 38/282 | 224 |
| teacher labels | +0.0046 | 40/282 | 242 |
| always no-trade | 0.0 | — | 0 |
Invalid JSON output rate: 0.0%. Action distribution: 94.6% no_trade.
The SFT adapter reproduces the teacher's format but not its profitability: on held-out playoff games it underperforms both the teacher and the no-trade baseline. The teacher's own edge is itself marginal (val P&L −0.0019/game). This is a working research pipeline, not a profitable trading policy. Sensible next steps: GRPO on the simulator reward, richer state features (news, lineup data), and higher-fidelity tick data (Betfair) if pursuing the edge seriously.
Prices are 1-minute-fidelity traded/mid prices without order-book depth; the replay assumes fills at displayed prices, which overstates achievable P&L. For research use only.