Update benchmarks/models/sansa/README.md (#6)
Browse files- Update benchmarks/models/sansa/README.md (eb2b2586d41846153dfb8d0fd7f63e9509fb5b21)
Co-authored-by: Taychinov Evgeny <TayJen@users.noreply.huggingface.co>
benchmarks/models/sansa/README.md
CHANGED
|
@@ -2,7 +2,7 @@
|
|
| 2 |
|
| 3 |
Training SANSA on Yambda-500M with Listens, on Yambda-5B with Likes and on Yambda-5B with Listens is infeasible due to the memory explosion during sparse matrix multiplication.
|
| 4 |
|
| 5 |
-
For example, given the base weights density of
|
| 6 |
|
| 7 |
One may observe that the good solution is to try to reduce the weights density. However given the results on Yambda-50M with Listens, further reducing density would collapse the model’s capacity, it becomes obvious that it is just practically futile.
|
| 8 |
|
|
|
|
| 2 |
|
| 3 |
Training SANSA on Yambda-500M with Listens, on Yambda-5B with Likes and on Yambda-5B with Listens is infeasible due to the memory explosion during sparse matrix multiplication.
|
| 4 |
|
| 5 |
+
For example, given the base weights density of \\(5e-5\\), the sparse matrix multiplication for Yambda-500M with Listens requires \\(466512103 * (3004578 ^ 2 * 5e-5) = 2.1e17\\) operations.
|
| 6 |
|
| 7 |
One may observe that the good solution is to try to reduce the weights density. However given the results on Yambda-50M with Listens, further reducing density would collapse the model’s capacity, it becomes obvious that it is just practically futile.
|
| 8 |
|