Llama3 problems
Complaints distilled from real conversations, 1,060 voices so far. Nothing here is invented.
WHAT PEOPLE COMPLAIN ABOUT
- Outages, errors, or limits get in the way (4 mentions)
- People find it expensive (3 mentions)
- Refuses or lectures more than people like (3 mentions)
- It can feel slow (2 mentions)
IN THEIR OWN WORDS
- SOURCE >
“[Bug] Llama 3.3-70B Galaxy prefill 2x slower at long context since 57180 (accurate SDPA exp on the fp32-dest path); prefix-caching benchmark red on every host.”
- SOURCE >
“llama3 always hallucinating?.”
- SOURCE >
“Since Llama 3 and Qwen3 have very large supported context sizes (131k and 41k tokens, respectively), the pre-allocated tensors consume ~8 GB of additional memory, which is quite expensive.”