NVFP4 experts for Blackwell, calibrated by NVIDIA's recipe and measured against the original: checkpoint plus the metrics.
AI & ML interests
Local Models Server and Free Ai Chat app
Recent Activity
View all activity
-
AtomicChat/Qwen3.8-27B-GGUF
Image-Text-to-Text • 27B • Updated • 306k • 168 -
AtomicChat/Qwen3.8-27B-GGUF-metrics
Updated • 8.39k • 4 -
AtomicChat/Qwen3.8-27B-MLX-AD-3.50bpw-DWQ-12.7GB
Image-Text-to-Text • 27B • Updated • 942 • 2 -
AtomicChat/Qwen3.8-27B-MLX-AD-3.70bpw-DWQ-13.4GB
Image-Text-to-Text • 27B • Updated • 923 • 3
Atomic Chat GGUF builds of Tencent Hy3 (295B MoE): coding-optimized imatrix quants
poolside Laguna XS 2.1 (33B/3B MoE) quantized by Atomic Chat - GGUF (llama.cpp) and MLX (Apple Silicon) ladders.
-
AtomicChat/Laguna-XS-2.1-GGUF
Text Generation • 33B • Updated • 599 • 14 -
AtomicChat/Laguna-XS-2.1-MLX-8bit
Text Generation • 33B • Updated • 377 • 2 -
AtomicChat/Laguna-XS-2.1-MLX-6bit
Text Generation • 33B • Updated • 193 • 2 -
AtomicChat/Laguna-XS-2.1-MLX-5bit
Text Generation • 33B • Updated • 475 • 3
Qwen2.5 instruct models, self-quantized to GGUF by Atomic Chat with a per-tensor importance matrix.
Atomic Chat self-quantized GGUFs of Liquid AI's LFM2.5 models, built from the original weights with an importance matrix.
Atomic Chat self-quantized GGUFs of Qwen3.5, built from the original weights with an importance matrix.
Atomic Chat GGUF builds of Google Gemma 4
Gemma 4 MTP assistant drafters as GGUF (F16/Q8_0/Q5_K_M/Q4_K_M/Q4_K_S). Speculative-decoding heads for the atomic-llama-cpp-turboquant fork.
-
AtomicChat/gemma-4-E2B-it-assistant-GGUF
Text Generation • 77.2M • Updated • 803 • 8 -
AtomicChat/gemma-4-E4B-it-assistant-GGUF
Text Generation • 78M • Updated • 1.36k • 23 -
AtomicChat/gemma-4-26B-A4B-it-assistant-GGUF
Text Generation • 0.4B • Updated • 2.29k • 29 -
AtomicChat/gemma-4-31B-it-assistant-GGUF
Text Generation • 0.5B • Updated • 2.44k • 20
Inkling by Thinking Machines, self-quantized to GGUF by Atomic Chat. 975B MoE (41B active), 1M context.
DFlash block-diffusion draft heads (GGUF). Attach to any target GGUF to speed up llama.cpp.
Microsoft Phi-4 models, self-quantized to GGUF by Atomic Chat with a per-tensor importance matrix.
DeepReinforce's Ornith 1.0 agentic-coding models, self-quantized to GGUF by Atomic Chat with a per-tensor importance matrix. Runs fully offline.
Atomic Chat self-quantized GGUFs of the Qwen3 Coder models, built from the original weights with an importance matrix.
Atomic Chat self-quantized GGUFs of Qwen3.6, built from the original weights with an importance matrix.
Dynamic-imatrix GGUF quants of Qwen 3.6 27B & 35B-A3B. TurboQuant3 KV + shared-model NextN ready.
NVFP4 experts for Blackwell, calibrated by NVIDIA's recipe and measured against the original: checkpoint plus the metrics.
-
AtomicChat/Qwen3.8-27B-GGUF
Image-Text-to-Text • 27B • Updated • 306k • 168 -
AtomicChat/Qwen3.8-27B-GGUF-metrics
Updated • 8.39k • 4 -
AtomicChat/Qwen3.8-27B-MLX-AD-3.50bpw-DWQ-12.7GB
Image-Text-to-Text • 27B • Updated • 942 • 2 -
AtomicChat/Qwen3.8-27B-MLX-AD-3.70bpw-DWQ-13.4GB
Image-Text-to-Text • 27B • Updated • 923 • 3
Inkling by Thinking Machines, self-quantized to GGUF by Atomic Chat. 975B MoE (41B active), 1M context.
Atomic Chat GGUF builds of Tencent Hy3 (295B MoE): coding-optimized imatrix quants
DFlash block-diffusion draft heads (GGUF). Attach to any target GGUF to speed up llama.cpp.
poolside Laguna XS 2.1 (33B/3B MoE) quantized by Atomic Chat - GGUF (llama.cpp) and MLX (Apple Silicon) ladders.
-
AtomicChat/Laguna-XS-2.1-GGUF
Text Generation • 33B • Updated • 599 • 14 -
AtomicChat/Laguna-XS-2.1-MLX-8bit
Text Generation • 33B • Updated • 377 • 2 -
AtomicChat/Laguna-XS-2.1-MLX-6bit
Text Generation • 33B • Updated • 193 • 2 -
AtomicChat/Laguna-XS-2.1-MLX-5bit
Text Generation • 33B • Updated • 475 • 3
Microsoft Phi-4 models, self-quantized to GGUF by Atomic Chat with a per-tensor importance matrix.
Qwen2.5 instruct models, self-quantized to GGUF by Atomic Chat with a per-tensor importance matrix.
DeepReinforce's Ornith 1.0 agentic-coding models, self-quantized to GGUF by Atomic Chat with a per-tensor importance matrix. Runs fully offline.
Atomic Chat self-quantized GGUFs of Liquid AI's LFM2.5 models, built from the original weights with an importance matrix.
Atomic Chat self-quantized GGUFs of the Qwen3 Coder models, built from the original weights with an importance matrix.
Atomic Chat self-quantized GGUFs of Qwen3.5, built from the original weights with an importance matrix.
Atomic Chat self-quantized GGUFs of Qwen3.6, built from the original weights with an importance matrix.
Atomic Chat GGUF builds of Google Gemma 4
Dynamic-imatrix GGUF quants of Qwen 3.6 27B & 35B-A3B. TurboQuant3 KV + shared-model NextN ready.
Gemma 4 MTP assistant drafters as GGUF (F16/Q8_0/Q5_K_M/Q4_K_M/Q4_K_S). Speculative-decoding heads for the atomic-llama-cpp-turboquant fork.
-
AtomicChat/gemma-4-E2B-it-assistant-GGUF
Text Generation • 77.2M • Updated • 803 • 8 -
AtomicChat/gemma-4-E4B-it-assistant-GGUF
Text Generation • 78M • Updated • 1.36k • 23 -
AtomicChat/gemma-4-26B-A4B-it-assistant-GGUF
Text Generation • 0.4B • Updated • 2.29k • 29 -
AtomicChat/gemma-4-31B-it-assistant-GGUF
Text Generation • 0.5B • Updated • 2.44k • 20