Qwen3-4B-Instruct-2507-REFFT

Hexagon HTP serving packs for Qwen3-4B-Instruct-2507.

1× and 2× are separate RFTB files. Both use pin-all single-image programs (retain=true, resident_window=0).

Qwen3-4B-Instruct-2507-w4-1xHTP.refft   # single HTP, no device collectives
Qwen3-4B-Instruct-2507-w4-2xHTP.refft   # dual HTP rank-sharded + 2 AR/layer

refft-hexagon cli --model Qwen3-4B-Instruct-2507-w4-1xHTP.refft \
  --world_size 1 --prompt "Who are you?" --max_new_tokens 12
refft-hexagon cli --model Qwen3-4B-Instruct-2507-w4-2xHTP.refft \
  --world_size 2 --prompt "Who are you?" --max_new_tokens 12

Measured on Dragonwing IQ-9075 with refft-hexagon 0.6.5.dev682+g9833e0505.

Downloads last month: -; Downloads are not tracked for this model. How to track