Repository navigation
Error with embed when passing a List #2199
Description
from llama_cpp import Llama
# This works
llm = Llama(model_path="/tmp/all-MiniLM-L6-v2-Q8_0.gguf", embedding=True)
print(llm.embed(["test"]))
# This also works
import numpy as np
print(np.array([llm.embed(x) for x in ["test", "test"]]))
# This fails
llm = Llama(model_path="/tmp/all-MiniLM-L6-v2-Q8_0.gguf", embedding=True)
print(llm.embed(["test", "test"]))Reacted by Mischa Lisovyi
Activity
fix(_internals): use n_tokens0 offset when enabling last-token logits…
mlisovyi commentedon May 5, 2026
Tests in #2205 seem to be failing with the same error as before (
llama_decode: failed to decode, ret = -1) . @abetlen is that expected?Related to #2051 (same error).
davidmezzetti commentedon May 5, 2026
Agree. That PR looked vibe coded.
davidmezzetti commentedon May 11, 2026
Thank you 🚀