Popular repositories Loading

1. ioteverythin-display ioteverythin-display Public

   Touch-i: A wall-mounted Home Assistant control panel on the Waveshare ESP32-S3-Touch-LCD-4. Switches, Climate, Media, Presence. HACS integration with zero-code setup.

   C 4 1

2. speculative-tools speculative-tools Public

   Speculative execution engine for LLM agent tool calls. Predicts and pre-executes tools in parallel to reduce agent latency.

   Python 1

3. prefixroute prefixroute Public

   Prefix cache affinity routing for multi-replica LLM serving. Routes each request to the replica already holding its prompt prefix, with a load guard against hotspots, cutting mean time to first tok…

   Python 1

4. kvprefetch kvprefetch Public

   Decode-latency planner for offloaded KV cache: double-buffered layer-ahead prefetch cuts Llama-3-8B decode step latency 37.7 percent at 2k context (17.6 to 11.0 ms) on an A100 PCIe card, and locate…

   Python 1

5. qmigrate qmigrate Public

   Per-layer activation-weight difficulty migration for int8 (W8A8) linear layers. An automatic per-layer migration-strength search cuts mean output quantization error 78.7 percent vs naive W8A8 and 1…

   Python 1

6. weather-clock weather-clock Public