From: Georgi Gerganov <redacted>
Date: Thu, 30 Mar 2023 21:34:14 +0000 (+0300)
Subject: gpt-2 : fix qunatize tool to quantize the "lm_head" tensor
X-Git-Tag: upstream/0.0.1642~1564
X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=87a91a5809b8014dedc2666d81246e86bdd9f8b0;p=pkg%2Fggml%2Fsources%2Fggml

gpt-2 : fix qunatize tool to quantize the "lm_head" tensor
---

diff --git a/examples/gpt-2/quantize.cpp b/examples/gpt-2/quantize.cpp
index 3cc48ea3..693b0787 100644
--- a/examples/gpt-2/quantize.cpp
+++ b/examples/gpt-2/quantize.cpp
@@ -162,6 +162,7 @@ bool gpt2_model_quantize(const std::string & fname_inp, const std::string & fnam
             // regexes of tensor names to be quantized
             const std::vector<std::string> k_names = {
                 "model/wte",
+                "model/lm_head",
                 "model/h.*/attn/c_attn/w",
                 "model/h.*/attn/c_proj/w",
                 "model/h.*/mlp/c_fc/w",