Jan Bernlöhr

jbernloehr

nvidia

4

·

janbernloehr

AI & ML interests

None yet

Recent Activity

new activity 13 days ago

Sandroeth/cali-0.1B:modeling_cali.py: bf16 forward crashes with "expected scalar type Float but found BFloat16" at torch.matmul(att, v) — float32 RoPE cache promotes q/k

new activity 14 days ago

ai-sage/GFusion-10B-A1.8B-base:IndexError in _extract_kv_cache on empty DynamicCache during first forward (prefill) with use_cache=True

new activity 19 days ago

ai-sage/GFusion-10B-A1.8B-base:IndexError in _extract_kv_cache on empty DynamicCache during first forward (prefill) with use_cache=True

View all activity

Organizations

New activity in Sandroeth/cali-0.1B 13 days ago

modeling_cali.py: bf16 forward crashes with "expected scalar type Float but found BFloat16" at torch.matmul(att, v) — float32 RoPE cache promotes q/k

#2 opened 13 days ago by

New activity in ai-sage/GFusion-10B-A1.8B-base 14 days ago

IndexError in _extract_kv_cache on empty DynamicCache during first forward (prefill) with use_cache=True

#1 opened 19 days ago by

New activity in vivo-ai/BlueLM-7B-Chat-32K 4 months ago

Fix softmax_scale type for torch.compile compatibility

#5 opened 4 months ago by

New activity in deepseek-ai/deepseek-moe-16b-base 5 months ago

Fix compatibility with transformers 5.0

#8 opened 5 months ago by