primitive-ai
/

Qwen3.8-Flash-Next-PLE-quant

qwen3.8-flash-next

Model card Files Files and versions

Qwen3.8-Flash-Next-PLE-quant

113 GB

Ctrl+K

Ctrl+K

2 contributors

History: 21 commits

jagat-primitive-org's picture

jagat-primitive-org

Card: the sidecar now also serves FP8-table checkpoints; nvidia's NVFP4 booted on one card with it

4f861b6 verified 8 days ago

assets
Add Primitive brand banner, logo and card thumbnail 16 days ago
ples_fp8
Upload folder using huggingface_hub 18 days ago
ples_int4
Upload folder using huggingface_hub 18 days ago
ples_nvfp4
Upload folder using huggingface_hub 17 days ago
.gitattributes

1.57 kB
Add Primitive brand banner, logo and card thumbnail 16 days ago
README.md

11.3 kB
Card: the sidecar now also serves FP8-table checkpoints; nvidia's NVFP4 booted on one card with it 8 days ago
connector_mrv2.py

19.2 kB
ship the async-MRV2 connector fix as an overlay (upstream 4e8b849) 14 days ago
ple_layer_quant.py

52.4 kB
Overlay: build the n-gram table parameter on the meta device when VLLM_PLE_QUANT_DIR is set, and swap the stub Parameter instead of set_data — removes the 102 GB virtual reservation that the kernel's overcommit heuristic refuses on hosts with less RAM+swap than the table (field report, 64 GB host); validated under an emulated 67/99 GiB commit limit, sanity PASS, tool-calling 77.0 (n=3), 81 tok/s c1 8 days ago
worker_image_quant.py

40.8 kB
Overlay: drop a checkpoint's own table tensors when a sidecar is attached, including the monolithic weight/weight_scale form — lets FP8-table checkpoints (nvidia, official FP8) serve on one card 8 days ago