SipsaLabs/phi-3.5-moe-instruct-uc-v3-bpw5
Updated
AI & ML interests
Lossless 5-bit transformer compression. SHA-256 verifiable bit-identical reconstruction across 22 architectures (dense + MoE + state-space). Hermes-3-Llama-3.1-405B at 1.0066x PPL on a single 32 GB GPU. OpenAI-compatible inference API at api.sipsalabs.com/v1.
Recent Activity
View all activity
SipsaLabs/mixtral-8x7b-v0.1-uc-v3-bpw5
Updated
SipsaLabs/yi-1.5-9b-uc-v3-bpw5
Updated
SipsaLabs/phi-3-mini-4k-instruct-uc-v3-bpw5
Updated
SipsaLabs/olmo-2-0425-1b-instruct-uc-v3-bpw5
Updated
SipsaLabs/smollm2-1.7b-instruct-uc-v3-bpw5
Updated
SipsaLabs/olmo-2-0425-1b-uc-v3-bpw5
Updated
SipsaLabs/tinyllama-1.1b-chat-v1.0-uc-v3-bpw5
Updated
SipsaLabs/smollm2-1.7b-uc-v3-bpw5
Updated
SipsaLabs/mixtral-8x22b-v0.1-uc-v3-bpw5
Updated
SipsaLabs/qwen3-235b-a22b-uc-v3-bpw5
Updated
SipsaLabs/llama-3.1-70b-uc-v3-bpw5
Updated
SipsaLabs/phi-3.5-moe-uc-v3-bpw5
Updated
SipsaLabs/qwen3-14b-uc-v3-bpw5
Updated
SipsaLabs/qwen3-8b-uc-v3-bpw5
Updated
SipsaLabs/qwen3-1.7b-uc-v3-bpw5
Updated
SipsaLabs/phi-3.5-moe-instruct-streaming-bpw5
Updated
SipsaLabs/mixtral-8x22b-v0.1-streaming-bpw5
Updated
SipsaLabs/mixtral-8x7b-v0.1-streaming-bpw5
Updated
SipsaLabs/qwen3-235b-a22b-streaming-bpw5
Updated
SipsaLabs/llama-3.1-8b-streaming-bpw5
Updated
SipsaLabs/llama-3.1-70b-streaming-bpw5
Updated
SipsaLabs/mistral-7b-v0.3-streaming-bpw5
Updated
SipsaLabs/qwen3-1.7b-streaming-bpw5
Updated
SipsaLabs/hermes-3-llama-3.1-405b-streaming-bpw5
Updated
SipsaLabs/qwen3-32b-streaming-bpw5
Updated
SipsaLabs/qwen2.5-72b-streaming-bpw5
Updated
SipsaLabs/qwen3-14b-streaming-bpw5
Updated
SipsaLabs/qwen3-1.7b-shared-block-preview
Updated
SipsaLabs/qwen3-8b-streaming-bpw5
Updated