diff options
| author | Kawrakow <iwankawrakow@gmail.com> | 2024-09-27 08:16:06 +0300 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2024-09-27 08:16:06 +0300 |
| commit | 6dec4af4b6e65eb72e646a6f8b10d77c9d306281 (patch) | |
| tree | b69a6dfdd024ccf6a4d7490666664cbac4bc65ce /src/llama.cpp | |
| parent | 546f3ef349a7082fbc349897c3c7246baed2a6c6 (diff) | |
Adding ability to have meta data per tensor row (#61)
* POC: per row scale
This is a POC how to work around opinionated ggml to
have scales per row rather than per block.
Only implemened for Zen4 and only for iq2_tn.
* POC per row scale: iq2_tn on NEON
* POC per row scale: iq2_tn on Metal
* Per row scale Metal templates
* iq1_tn: shrink to 1.625 bpw (NEON and Metal)
* POC per row scale: CUDA
* POC per row scale: add CUDA TODOs
There are two places in ggml-cuda.cu left where it is assumed
that type_size * n_per_row / block_size is the way to compute
and handle row sizes. This does not affect simple usage,
but will lead to issues when tensors are split between GPUs.
* Per row scales - CUDA
The only place left where there are unnecessary assumptions being made
is in the Flash Attention code. As we are not using any quants that
use per row scales for quantized KV cache, it should be OK for now.
* Update IQ1_TN and IQ2_TN bpw shown to user
---------
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
Diffstat (limited to 'src/llama.cpp')
| -rw-r--r-- | src/llama.cpp | 4 |
1 files changed, 2 insertions, 2 deletions
diff --git a/src/llama.cpp b/src/llama.cpp index 0eea948a..df57c071 100644 --- a/src/llama.cpp +++ b/src/llama.cpp @@ -4498,9 +4498,9 @@ static std::string llama_model_ftype_name(llama_ftype ftype) { case LLAMA_FTYPE_MOSTLY_IQ5_K: return "IQ5_K - 5.5 bpw"; case LLAMA_FTYPE_MOSTLY_IQ6_K: return "IQ6_K - 6.6 bpw"; case LLAMA_FTYPE_MOSTLY_IQ1_BN: return "IQ1_BN - 1.625 bpw Bitnet"; - case LLAMA_FTYPE_MOSTLY_IQ1_TN: return "IQ1_TN - 1.6875 bpw TriLM"; + case LLAMA_FTYPE_MOSTLY_IQ1_TN: return "IQ1_TN - 1.625 bpw TriLM"; case LLAMA_FTYPE_MOSTLY_IQ2_BN: return "IQ2_BN - 2.00 bpw Bitnet"; - case LLAMA_FTYPE_MOSTLY_IQ2_TN: return "IQ2_TN - 2.06 bpw TriLM"; + case LLAMA_FTYPE_MOSTLY_IQ2_TN: return "IQ2_TN - 2.00 bpw TriLM"; case LLAMA_FTYPE_MOSTLY_IQ3_S: return "IQ3_S - 3.4375 bpw"; case LLAMA_FTYPE_MOSTLY_IQ3_M: return "IQ3_S mix - 3.66 bpw"; case LLAMA_FTYPE_MOSTLY_Q4_0_4_4: return "Q4_0_4_4"; |