SnoopyFan commited on
Commit
01523d8
·
verified ·
1 Parent(s): 2f65800

Add WorldCompass Sana training artifacts

Browse files
rebuttal_data/worldcompass_sana/wc_sana_formal_local8_b979afd_20260801210826/node_1/services/hpsv3.log ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.52, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.
2
+ You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
3
+
4
+ Some weights of Qwen2VLRewardModelBT were not initialized from the model checkpoint at /mnt/workspace/fly/hf-cache/hub/models--Qwen--Qwen2-VL-7B-Instruct/snapshots/eed13092ef92e448dd6875b2a00151bd3f7db0ac and are newly initialized: ['rm_head.0.bias', 'rm_head.0.weight', 'rm_head.3.bias', 'rm_head.3.weight', 'rm_head.5.bias', 'rm_head.5.weight']
5
+ You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.
6
+ [serve_hpsv3] ready model=MizzenAI/HPSv3@4f81e3e09edd82fe3c5f636444c721b592a735ca sha256=a13d7ff5a07b7ffa0f7824e60d62e6ae144541ceefd5224b4c08fda7ab39f353 base=Qwen/Qwen2-VL-7B-Instruct@eed13092ef92e448dd6875b2a00151bd3f7db0ac device=cuda:0 address=('127.0.0.1', 18281)