speed is very slow #28
Comments
|
Running on a GPU, I find that dumping extracted features takes up most time. So you may optimize it yourself. |
|
Hi, these examples are provided as starting point to write your own training scripts using the package modules. I don't plan to update them any further. |
jameshennessytempus
pushed a commit
to jameshennessytempus/transformers
that referenced
this issue
Jun 1, 2023
1 task
ocavue
pushed a commit
to ocavue/transformers
that referenced
this issue
Sep 13, 2023
jonb377
pushed a commit
to jonb377/hf-transformers
that referenced
this issue
Nov 3, 2023
Summary: This pull requst tries to shard every matmul in LLaMA, below is the sharding strategy: 1. up_proj (batch, length, intermediate): mesh (data, None, model) 2. gate_proj (batch, length, intermediate): mesh (data, None, model) 3. down_proj (batch, length, hidden): mesh (data, None, model) 4. query_states (batch, length, hidden): mesh (data, None, model) 5. key_states (batch, length, hidden / attention_heads * key_value_heads): mesh (data, None, model) 6. value_states (batch, length, hidden / attention_heads * key_value_heads): mesh (data, None, model) 7. attn_weights (batch, num_attention_heads, length, length): mesh (data, model, none, none) 8. attn_output (batch, length, hidden): mesh (data, None, model) 9. hidden_states (batch, length, hidden): mesh (data, None, model) Test Plan: Tested on v4-8
SangbumChoi
added a commit
to SangbumChoi/transformers
that referenced
this issue
Jan 13, 2025
Add: yolov6{m,l}
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
convert samples to features, is very slow
The text was updated successfully, but these errors were encountered: