Commit Graph

517 Commits

Author SHA1 Message Date
ed4d0959d3 Add a const to easily tweak the dtype used for llama internal computations. 2023-06-30 15:01:39 +01:00
fbc329ed85 Add the verbose cpu cast operations. 2023-06-30 10:33:29 +01:00
8ad47907f3 Add the kernels. 2023-06-30 10:26:56 +01:00
19cbbc5212 Improve how we check that the dims are in bounds. 2023-06-30 09:11:00 +01:00
f6152e74b6 Tweak the kv-cache flag. 2023-06-29 22:16:40 +01:00
ae3f202f3b Add a flag. 2023-06-29 22:12:15 +01:00
23389b1bd7 Enable the KV cache after fixing the caching length and the rope bits. 2023-06-29 22:00:57 +01:00
b50bd880ce Only narrow when needed + deactivate the kv cache. 2023-06-29 19:07:52 +01:00
3232df9458 Add some KV cache to llama. 2023-06-29 15:29:40 +01:00
889f7e0971 Merge pull request #39 from LaurentMazare/anyhow-backtrace
Add backtraces.
2023-06-29 13:17:53 +01:00
e27ee98d3f Add backtraces. 2023-06-29 13:17:20 +01:00
78ec40b077 Typo. 2023-06-29 12:09:53 +00:00
de48e6fd59 Putting back main. 2023-06-29 12:08:35 +00:00
0958c588f6 Putting back seed. 2023-06-29 12:07:21 +00:00
c5e8f788be Revert some changes. 2023-06-29 12:05:53 +00:00
e63ed6aaa3 Remove unwrap. 2023-06-29 12:04:25 +00:00
2fe1d3e36d Moving llama to f16. 2023-06-29 12:00:16 +00:00
b4dc9f6108 Add a seed parameter to llama. 2023-06-29 12:47:19 +01:00
53628db3a9 Merge pull request #36 from LaurentMazare/fix_example
Simple example fix.
2023-06-29 13:36:05 +02:00
1913512f42 Simple example fix. 2023-06-29 11:10:57 +00:00
2741b39ad3 Use broadcasted scalars for const tensors. 2023-06-29 11:56:40 +01:00
3872dc4751 Merge pull request #19 from LaurentMazare/llama_safetensors
Llama safetensors
2023-06-29 12:49:26 +02:00
b4aab7b95f Put more requirements on the withdtype trait. 2023-06-29 11:37:42 +01:00
c9c468e1aa Use Map2 for binary ops. 2023-06-29 10:09:15 +01:00
83c7d660ca Add Map2. 2023-06-29 10:05:06 +01:00
367170da45 Also use Map1 for embedding. 2023-06-29 09:45:27 +01:00
8ad03a5fb6 Use Map1 on unary ops. 2023-06-29 09:37:38 +01:00
fff13dbb4e Factorize the kernel naming scheme. 2023-06-29 09:29:59 +01:00
d3c7b0d168 Use Map1 for sum. 2023-06-29 09:27:07 +01:00
122e334d0c Simplify the pattern matching logic in the cuda backend. 2023-06-29 09:21:11 +01:00
eaa3ce359e Cosmetic change. 2023-06-28 22:02:23 +01:00
1328b5cb20 Factor some code out. 2023-06-28 21:56:44 +01:00
c583ee0f2c Add map2. 2023-06-28 21:38:01 +01:00
46c07b924c Tweak some comment. 2023-06-28 21:10:54 +01:00
2ae368e98e Switch from a macro to a trait to make things more generic. 2023-06-28 21:06:56 +01:00
ece3ec6167 Final updates -> moving to deterministic for easier comparison. 2023-06-28 14:56:39 +00:00
926fffa0b7 Ok. 2023-06-28 14:56:39 +00:00
e29dae044d Tmp. 2023-06-28 14:56:38 +00:00
6c9e6b5a99 Get the cuda tests to pass. 2023-06-28 15:53:23 +01:00
3f0d9fbb25 Adapt the cuda bits. 2023-06-28 15:43:03 +01:00
cca699be6c Fix some cpu issue. 2023-06-28 15:09:15 +01:00
1c755c0e5b Remove some todos. 2023-06-28 14:33:06 +01:00
caafef6cc1 Get the cpu tests to run. 2023-06-28 14:32:02 +01:00
14449ff80c Get the cpu backend to compile. 2023-06-28 14:12:38 +01:00
54a6c40f27 Propagate the changes on the cpu backend. 2023-06-28 14:00:49 +01:00
303b853098 Propagate the layout refactoring. 2023-06-28 13:42:23 +01:00
30b355ccd2 Simplify the narrow implementation. 2023-06-28 13:09:59 +01:00
c1bbbf94f6 Start refactoring the stride. 2023-06-28 12:57:30 +01:00
7938d2b848 Add the grad for narrow. 2023-06-28 10:46:00 +01:00
615196e7be Add more gradients. 2023-06-28 09:59:52 +01:00