1-bit inference of 0.8m param gpt running inside 8192 bytes of sram
00:00
(video is 100x speedup)
1-bit weights stored in flash; int16 activations; int8 kv cache
- @steipete
we can hook up openclaw to this right
FPGA time
1-bit inference of 0.8m param gpt running inside 8192 bytes of sram
00:00
(video is 100x speedup)
1-bit weights stored in flash; int16 activations; int8 kv cache
we can hook up openclaw to this right
FPGA time