Qwen3-1.7B-REFFT

Hexagon HTP serving bundles for Qwen3-1.7B.

Each artifact ending in .refft is one packaging variant for refft-hexagon (cli | serve).

Preferred layout (single-file RFTB)

Qwen3-1.7B-w4-1xHTP.refft   # single HTP, pin-all, no collectives
Qwen3-1.7B-w4-2xHTP.refft   # dual HTP rank-sharded + 2 AR/layer

1× and 2× are separate RFTBs (do not mix). Both are pin-all single-image programs (retain=true, resident_window=0).

  • Container: scheme-driven RFTB (general.scheme=hexagon.serving.v1)
  • Load: mmap BlobStore (no extract)
  • Runtime: generic Visa LM path (host_ops.embed.placement=host_only, multi weight-page pin-all)
refft-hexagon cli --model Qwen3-1.7B-w4-1xHTP.refft \
  --world_size 1 --prompt "Who are you?" --max_new_tokens 12
refft-hexagon cli --model Qwen3-1.7B-w4-2xHTP.refft \
  --world_size 2 --prompt "Who are you?" --max_new_tokens 12

See: docs/guides/refft_serving_bundle_rftb.md, docs/guides/refft_generic_execution_contract.md, docs/guides/refft_serving_pack_test_upload.md.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support