huggingface/text-generation-inference
5
0
Fork 0
mirror of https://github.com/huggingface/text-generation-inference.git synced 2025-06-19 15:52:08 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fix batching

Browse Source 
Signed-off-by: Adrien Gallouët <angt@huggingface.co>
This commit is contained in:
Adrien Gallouët 2025-02-01 16:09:51 +00:00
parent 2a51e415ff
commit 96434a1e7e
No known key found for this signature in database
1 changed files with 6 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
backends/llamacpp/src/backend.rs
View File

@ -429,12 +429,15 @@ impl LlamacppBackend {
requests = Vec::new(); requests = Vec::new();
continue; continue;
} }
if n_tokens + request.input_ids.len() > conf.max_batch_total_tokens as usize { let n_tokens_to_add = request.input_ids.len();
if n_tokens + n_tokens_to_add > conf.max_batch_total_tokens as usize {
let _ = sync_tx.send(requests); let _ = sync_tx.send(requests);
n_tokens = request.input_ids.len(); n_tokens = n_tokens_to_add;
requests = vec![request]; requests = vec![request];
continue; continue;
} }
n_tokens += n_tokens_to_add;
requests.push(request); requests.push(request);
}, },
Err(_) => { Err(_) => {
@ -487,7 +490,7 @@ impl LlamacppBackend {
seqs.push(LlamacppSeq { seqs.push(LlamacppSeq {
id: seq_id, id: seq_id,
batch_pos: llamacpp.batch.n_tokens as usize - 1, batch_pos: llamacpp.batch.n_tokens as usize - 1,
token: -1, token: bindings::LLAMA_TOKEN_NULL,
pos: last_pos as bindings::llama_pos + 1, pos: last_pos as bindings::llama_pos + 1,
sampler: sampler, sampler: sampler,
text: String::with_capacity(1024), text: String::with_capacity(1024),