← Back to context Comment by elvircrn 4 days ago Good stuff. Having fp8/fp4 would be great, too! 4 comments elvircrn Reply lifthrasiir 4 days ago There are too many fp8/fp4 formats out there, many with different trade-offs. At least fp16 is standardized... elvircrn 4 days ago Yes, all of them would be great. lifthrasiir 3 days ago Hmm, you successfully nerd-sniped me. AFAIK there are at least:- 8-bit floating point types: (S1)E4M3 (finite only, has no-negative-zero variant), (S1)E5M2 (has no-negative-zero variant), E8M0- 6-bit floating point types: (S1)E2M3, (S1)E3M2- 4-bit floating point types: (S1)E2M1, NF4- Block floating point types: MXFP8 (E5M2/E4M3*32+E8M0), MXFP6 (E3M2/E2M3*32+E8M0), MXFP4 (E2M1*32+E8M0), NXFP [1][1] https://arxiv.org/pdf/2412.19821Do you have anything to add on? 1 reply →
lifthrasiir 4 days ago There are too many fp8/fp4 formats out there, many with different trade-offs. At least fp16 is standardized... elvircrn 4 days ago Yes, all of them would be great. lifthrasiir 3 days ago Hmm, you successfully nerd-sniped me. AFAIK there are at least:- 8-bit floating point types: (S1)E4M3 (finite only, has no-negative-zero variant), (S1)E5M2 (has no-negative-zero variant), E8M0- 6-bit floating point types: (S1)E2M3, (S1)E3M2- 4-bit floating point types: (S1)E2M1, NF4- Block floating point types: MXFP8 (E5M2/E4M3*32+E8M0), MXFP6 (E3M2/E2M3*32+E8M0), MXFP4 (E2M1*32+E8M0), NXFP [1][1] https://arxiv.org/pdf/2412.19821Do you have anything to add on? 1 reply →
elvircrn 4 days ago Yes, all of them would be great. lifthrasiir 3 days ago Hmm, you successfully nerd-sniped me. AFAIK there are at least:- 8-bit floating point types: (S1)E4M3 (finite only, has no-negative-zero variant), (S1)E5M2 (has no-negative-zero variant), E8M0- 6-bit floating point types: (S1)E2M3, (S1)E3M2- 4-bit floating point types: (S1)E2M1, NF4- Block floating point types: MXFP8 (E5M2/E4M3*32+E8M0), MXFP6 (E3M2/E2M3*32+E8M0), MXFP4 (E2M1*32+E8M0), NXFP [1][1] https://arxiv.org/pdf/2412.19821Do you have anything to add on? 1 reply →
lifthrasiir 3 days ago Hmm, you successfully nerd-sniped me. AFAIK there are at least:- 8-bit floating point types: (S1)E4M3 (finite only, has no-negative-zero variant), (S1)E5M2 (has no-negative-zero variant), E8M0- 6-bit floating point types: (S1)E2M3, (S1)E3M2- 4-bit floating point types: (S1)E2M1, NF4- Block floating point types: MXFP8 (E5M2/E4M3*32+E8M0), MXFP6 (E3M2/E2M3*32+E8M0), MXFP4 (E2M1*32+E8M0), NXFP [1][1] https://arxiv.org/pdf/2412.19821Do you have anything to add on? 1 reply →
There are too many fp8/fp4 formats out there, many with different trade-offs. At least fp16 is standardized...
Yes, all of them would be great.
Hmm, you successfully nerd-sniped me. AFAIK there are at least:
- 8-bit floating point types: (S1)E4M3 (finite only, has no-negative-zero variant), (S1)E5M2 (has no-negative-zero variant), E8M0
- 6-bit floating point types: (S1)E2M3, (S1)E3M2
- 4-bit floating point types: (S1)E2M1, NF4
- Block floating point types: MXFP8 (E5M2/E4M3*32+E8M0), MXFP6 (E3M2/E2M3*32+E8M0), MXFP4 (E2M1*32+E8M0), NXFP [1]
[1] https://arxiv.org/pdf/2412.19821
Do you have anything to add on?
1 reply →