Qwen3.5-9B General Game
General-game training run gg-q35-256g-b256-s42-20260911e.
main contains final checkpoint-4183. Use a checkpoint branch explicitly for reproducibility,
and pass the same revision to both model and processor/tokenizer loading.
from transformers import AutoProcessor, Qwen3_5ForConditionalGeneration
repo = "ltzheng/Qwen3.5-9B-General-Game"
revision = "checkpoint-1800"
processor = AutoProcessor.from_pretrained(repo, revision=revision)
model = Qwen3_5ForConditionalGeneration.from_pretrained(repo, revision=revision, dtype="auto", device_map="auto")
The tokenizer includes the five training action/thought tokens (IDs 248077–248081).
The chat template does not inject a thinking prefix. Use skip_special_tokens=False
when decoding outputs that must retain the game protocol delimiters.
These are training checkpoints; conversion/inference checks do not measure game performance.
Checkpoints
| Step | Branch | Verified commit |
|---|---|---|
| 100 | checkpoint-100 | 659733f9a23f4da8fe6c369ebfd35f8df5d8bb3c |
| 200 | checkpoint-200 | e656477b7d8502e18663563baed04ce4892b9a65 |
| 300 | checkpoint-300 | 4894b697cbdcb138b92934d06ef200792d455f19 |
| 400 | checkpoint-400 | 8680ea7e501a8686bc50898f066099ef951593bd |
| 500 | checkpoint-500 | 41f807862b78b17138ea36138dbc14bdf2abf575 |
| 600 | checkpoint-600 | accc471730c4046ce935c136b4fd360e31022457 |
| 700 | checkpoint-700 | 7127d0d34b75f8760db10626a4c7c0c0c2d00fd0 |
| 800 | checkpoint-800 | 9acb92d9555f5830dc3f6b81665f6d9c8a53372e |
| 900 | checkpoint-900 | 9836dcdd689c69628dd8a6fd1b4e5140b33289ed |
| 1000 | checkpoint-1000 | e2f11cc36b830fb2f5ee8d4205843746d3f013e8 |
| 1100 | checkpoint-1100 | a638e23586c39d57ee571d633c0f269141339870 |
| 1200 | checkpoint-1200 | 8513c89b20cf25aef222cf8f80c6b1bc1b89e07c |
| 1300 | checkpoint-1300 | 94dcfae07ce93beea95b513cd3ad0fb9844256da |
| 1400 | checkpoint-1400 | 145156c33c29ac87c3d1d871c04f7d1cfbad0c4e |
| 1500 | checkpoint-1500 | 12ae2dfd8d0c88511dd000be7a12b5ca91d75a72 |
| 1600 | checkpoint-1600 | 7b858402938b9ea4fee63ca918f0a67a98d2b3e1 |
| 1700 | checkpoint-1700 | 7eae5a2ba6d170a7c3d39ccf444d18398d61cb53 |
| 1800 | checkpoint-1800 | 20945c1a001ee98ae7abdfa3b7c3a93c1759fe89 |
| 1900 | checkpoint-1900 | 84f81c6941282a011695810e199d2da0b6166adb |
| 2000 | checkpoint-2000 | 2ac0c96f9c7df5312acd3f295ab45fb26402230e |
| 2100 | checkpoint-2100 | da9f847233050fdecc70283e6ee5d9006a2c3e41 |
| 2200 | checkpoint-2200 | e2ef892ed8c1d7f6dacdcf12984c5859a411f00a |
| 2300 | checkpoint-2300 | 43c43d79b8a7e8a7d79c787971a0a1b4dba34da9 |
| 2400 | checkpoint-2400 | 7f4a7580d8935589a94f5a281eeb3b7c4d6eebfc |
| 2500 | checkpoint-2500 | ad776d742ba6dbc38a35b8b34ae7b69a8c2a9ca9 |
| 2600 | checkpoint-2600 | 394343c0998c28771c45e5aaef30824222eec596 |
| 2700 | checkpoint-2700 | 490b460d977183df5077d8c2aef5a6ebc5363f71 |
| 2800 | checkpoint-2800 | 6c82a4c0b73ad9cb6da7d7e331c2ff41b7a620a3 |
| 2900 | checkpoint-2900 | 07ebdaab152ebe90600a0e7575bc41d11bf23fb8 |
| 3000 | checkpoint-3000 | fe338318516fa53ea2a317d7a8fda9cdf2a37d28 |
| 3100 | checkpoint-3100 | 15fa34e7eec80d067889f206d17727a2f43431e1 |
| 3200 | checkpoint-3200 | 7376a14f46a3bece1e91e28bf60430f353baf332 |
| 3300 | checkpoint-3300 | 4a20274b3ed847a155a7f2abf672ab38b80807be |
| 3400 | checkpoint-3400 | 0e9f927979def7bea6519de00a7991e3791091e0 |
| 3500 | checkpoint-3500 | 6db5e8da39d4314ca25a8200c2b4f778173c745e |
| 3600 | checkpoint-3600 | 54fef1e2b4d2cf24403c8b95aca57bbfae1a28b5 |
| 3700 | checkpoint-3700 | 477ccb072a7ee34276901955324b17628f2b745d |
| 3800 | checkpoint-3800 | d6f526ea07ecfce52929aa5df083d053600d20a2 |
| 3900 | checkpoint-3900 | efd58c375687114b134165440ff584652fc09363 |
| 4000 | checkpoint-4000 | 3f9d2691f705749a663b3681501af11abb0ece4e |
| 4100 | checkpoint-4100 | 399c484cef97f1da65ee8cd2eeef4668c35e5acd |
| 4183 | checkpoint-4183 | 58e9452fb0a89d226690e258593678e10c153147 |