Metal GGML Q40 support #1643
Merged
Metal GGML Q40 support #1643
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
LouisChourakiSonos
force-pushed
the
metal_ggml_matmul
branch
from
b7613ba to
7ce03b7
Compare
This reverts commit 6af0994.
kali
approved these changes
Feb 14, 2025
| && matches!(facts[0].datum_type, DatumType::F16 | DatumType::F32)) | ||
| || ((facts[1].datum_type == DatumType::F32) && (facts[0].datum_type == DatumType::F32)) | ||
| || ((facts[1].datum_type == DatumType::F16) | ||
| && matches!(facts[0].datum_type, DatumType::F32 | DatumType::F16)) |
There was a problem hiding this comment.
lol, this is very hard to parse :) maybe we could split the tricky q40 case apart then do somethings like
use DatumType::*;
match (facts[0].datum_type, facts[1].datum_type) {
(F32, F32) => true,
(F16, F16) => true,
(F16, F32) => true,
_ => false
}
| a_dt: DatumType, | ||
| b_dt: DatumType, | ||
| ) -> TractResult<TVec<TypedFact>> { | ||
| let out_dt = self.matmul.output_dt(a_dt, b_dt)?; |
There was a problem hiding this comment.
how about something that respect the symmetry more ?
ensure!([DatumType::F16, DatumType::F32].contains(out_dt));
Ok(tvec!(out_dt.fact(shape))
metal/src/kernels/matmul/mod.rs
Outdated
| }; | ||
|
|
||
| let mut rng = rand::thread_rng(); | ||
| let lhs_data: Vec<F> = (0..b * m * k) // Create a vector with 10 elements |
There was a problem hiding this comment.
i suspect we have a lying comment here.
metal/src/kernels/matmul/mod.rs
Outdated
|
|
||
| let mut rng = rand::thread_rng(); | ||
| let lhs_data: Vec<F> = (0..b * m * k) // Create a vector with 10 elements | ||
| .map(|_| F::from(rng.gen_range(0.0..1.0)).unwrap()) // Generate a random float in [0.0, 1.0) |
There was a problem hiding this comment.
mmm... this is not good. we should not rand() ourselves, but let the proptest strategies do it in predictable and simplifiable manner. The idea is to use proptest::collections::vec() or something like that.
LouisChourakiSonos
force-pushed
the
ggml_Q4
branch
from
ee3db0e to
1a5de36
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
No description provided.