rdtand/Qwen3.5-122B-A10B-PrismaQuant-4.75bit-vllm at main

Qwen3.5-122B-A10B-PrismaQuant-4.75bit-vllm

76.7 GB

Ctrl+K

1 contributor

History: 10 commits

rdtand

README: document scale_sweep polish (closed-form AutoRound analog) + bake-off table

0f226dc verified about 2 months ago

.gitattributes

1.64 kB
Add files using upload-large-folder tool 2 months ago
README.md

11 kB
README: document scale_sweep polish (closed-form AutoRound analog) + bake-off table about 2 months ago
chat_template.jinja

7.76 kB
Add files using upload-large-folder tool 2 months ago
config.json

57.8 kB
PrismaQuant final: visual NVFP4 (108/110 DP-placed), lm_head BF16 (vLLM ParallelLMHead runtime limit), pos_embed excluded about 2 months ago
generation_config.json

244 Bytes
PrismaQuant Phase 2 incremental: body AWQ-safe GPTQ+act-round, visual Fisher-allocated (108/111 Linears via per-Linear DP) about 2 months ago
merges.txt

3.35 MB
PrismaQuant Phase 2 incremental: body AWQ-safe GPTQ+act-round, visual Fisher-allocated (108/111 Linears via per-Linear DP) about 2 months ago
mixed_native_manifest.json

560 Bytes
PrismaQuant final: visual NVFP4 (108/110 DP-placed), lm_head BF16 (vLLM ParallelLMHead runtime limit), pos_embed excluded about 2 months ago
model-00001-of-00016.safetensors

5.11 GB
xet

PrismaQuant: GPTQ + closed-form scale_sweep polish (replaces act_round). Measured ~67% per-Linear output-MSE reduction vs prior pipeline in bake-off. about 2 months ago
model-00002-of-00016.safetensors

5.11 GB
xet

PrismaQuant: GPTQ + closed-form scale_sweep polish (replaces act_round). Measured ~67% per-Linear output-MSE reduction vs prior pipeline in bake-off. about 2 months ago
model-00003-of-00016.safetensors

5.11 GB
xet

PrismaQuant: GPTQ + closed-form scale_sweep polish (replaces act_round). Measured ~67% per-Linear output-MSE reduction vs prior pipeline in bake-off. about 2 months ago
model-00004-of-00016.safetensors

5.09 GB
xet

PrismaQuant: GPTQ + closed-form scale_sweep polish (replaces act_round). Measured ~67% per-Linear output-MSE reduction vs prior pipeline in bake-off. about 2 months ago
model-00005-of-00016.safetensors

5.11 GB
xet

PrismaQuant: GPTQ + closed-form scale_sweep polish (replaces act_round). Measured ~67% per-Linear output-MSE reduction vs prior pipeline in bake-off. about 2 months ago
model-00006-of-00016.safetensors

5.11 GB
xet

PrismaQuant final: visual NVFP4 (108/110 DP-placed), lm_head BF16 (vLLM ParallelLMHead runtime limit), pos_embed excluded about 2 months ago
model-00007-of-00016.safetensors

5.1 GB
xet

PrismaQuant final: visual NVFP4 (108/110 DP-placed), lm_head BF16 (vLLM ParallelLMHead runtime limit), pos_embed excluded about 2 months ago
model-00008-of-00016.safetensors

5.11 GB
xet

PrismaQuant final: visual NVFP4 (108/110 DP-placed), lm_head BF16 (vLLM ParallelLMHead runtime limit), pos_embed excluded about 2 months ago
model-00009-of-00016.safetensors

5.11 GB
xet

PrismaQuant: GPTQ + closed-form scale_sweep polish (replaces act_round). Measured ~67% per-Linear output-MSE reduction vs prior pipeline in bake-off. about 2 months ago
model-00010-of-00016.safetensors

5.11 GB
xet

PrismaQuant final: visual NVFP4 (108/110 DP-placed), lm_head BF16 (vLLM ParallelLMHead runtime limit), pos_embed excluded about 2 months ago
model-00011-of-00016.safetensors

5.11 GB
xet

PrismaQuant: GPTQ + closed-form scale_sweep polish (replaces act_round). Measured ~67% per-Linear output-MSE reduction vs prior pipeline in bake-off. about 2 months ago
model-00012-of-00016.safetensors

5.11 GB
xet

PrismaQuant: GPTQ + closed-form scale_sweep polish (replaces act_round). Measured ~67% per-Linear output-MSE reduction vs prior pipeline in bake-off. about 2 months ago
model-00013-of-00016.safetensors

5.11 GB
xet

PrismaQuant final: visual NVFP4 (108/110 DP-placed), lm_head BF16 (vLLM ParallelLMHead runtime limit), pos_embed excluded about 2 months ago
model-00014-of-00016.safetensors

5.11 GB
xet

PrismaQuant: GPTQ + closed-form scale_sweep polish (replaces act_round). Measured ~67% per-Linear output-MSE reduction vs prior pipeline in bake-off. about 2 months ago
model-00015-of-00016.safetensors

5.04 GB
xet

PrismaQuant: GPTQ + closed-form scale_sweep polish (replaces act_round). Measured ~67% per-Linear output-MSE reduction vs prior pipeline in bake-off. about 2 months ago
model-00016-of-00016.safetensors

104 MB
xet

PrismaQuant Phase 2 incremental: body AWQ-safe GPTQ+act-round, visual Fisher-allocated (108/111 Linears via per-Linear DP) about 2 months ago
model.safetensors.index.json

17.4 MB
xet

PrismaQuant final: visual NVFP4 (108/110 DP-placed), lm_head BF16 (vLLM ParallelLMHead runtime limit), pos_embed excluded about 2 months ago
preprocessor_config.json

390 Bytes
PrismaQuant Phase 2 incremental: body AWQ-safe GPTQ+act-round, visual Fisher-allocated (108/111 Linears via per-Linear DP) about 2 months ago
processor_config.json

1.19 kB
Add files using upload-large-folder tool 2 months ago
tokenizer.json

12.8 MB
xet

PrismaQuant Phase 2 incremental: body AWQ-safe GPTQ+act-round, visual Fisher-allocated (108/111 Linears via per-Linear DP) about 2 months ago
tokenizer_config.json

16.7 kB
PrismaQuant Phase 2 incremental: body AWQ-safe GPTQ+act-round, visual Fisher-allocated (108/111 Linears via per-Linear DP) about 2 months ago
video_preprocessor_config.json

385 Bytes
PrismaQuant Phase 2 incremental: body AWQ-safe GPTQ+act-round, visual Fisher-allocated (108/111 Linears via per-Linear DP) about 2 months ago
vocab.json

6.72 MB
PrismaQuant Phase 2 incremental: body AWQ-safe GPTQ+act-round, visual Fisher-allocated (108/111 Linears via per-Linear DP) about 2 months ago