Remove deprecated rules for token embeddings

Nexesenex · Nexesenex · commit 53b8eaa316e8 · 2024-08-25T03:12:29.000+02:00
diff --git a/src/llama.cpp b/src/llama.cpp
@@ -16514,9 +16514,6 @@ static ggml_type llama_tensor_get_type(quantize_state_internal & qs, ggml_type n
                 if (qs.model.hparams.n_vocab >= 127999 || qs.model.hparams.n_head() <= 20) new_type = GGML_TYPE_IQ3_S;
                 new_type = GGML_TYPE_IQ4_XS;
             }
-            else if (ftype == LLAMA_FTYPE_MOSTLY_IQ2_S || ftype == LLAMA_FTYPE_MOSTLY_IQ2_M) new_type = GGML_TYPE_IQ3_XXS;
-            else if (ftype == LLAMA_FTYPE_MOSTLY_IQ2_XL) new_type = GGML_TYPE_IQ3_XXS;
-            else if (ftype == LLAMA_FTYPE_MOSTLY_IQ3_XXS) new_type = GGML_TYPE_IQ3_S;
             else if (new_type == GGML_TYPE_Q4_0_4_4 || new_type == GGML_TYPE_Q4_0_4_8 || new_type == GGML_TYPE_Q4_0_8_8) {
                 new_type = GGML_TYPE_Q4_0;
             }

Original file line number	Diff line number	Diff line change
`@@ -16514,9 +16514,6 @@ static ggml_type llama_tensor_get_type(quantize_state_internal & qs, ggml_type n`
`16514`	`16514`	`if (qs.model.hparams.n_vocab >= 127999 \|\| qs.model.hparams.n_head() <= 20) new_type = GGML_TYPE_IQ3_S;`
`16515`	`16515`	`new_type = GGML_TYPE_IQ4_XS;`
`16516`	`16516`	`}`
`16517`		`- else if (ftype == LLAMA_FTYPE_MOSTLY_IQ2_S \|\| ftype == LLAMA_FTYPE_MOSTLY_IQ2_M) new_type = GGML_TYPE_IQ3_XXS;`
`16518`		`- else if (ftype == LLAMA_FTYPE_MOSTLY_IQ2_XL) new_type = GGML_TYPE_IQ3_XXS;`
`16519`		`- else if (ftype == LLAMA_FTYPE_MOSTLY_IQ3_XXS) new_type = GGML_TYPE_IQ3_S;`
`16520`	`16517`	`else if (new_type == GGML_TYPE_Q4_0_4_4 \|\| new_type == GGML_TYPE_Q4_0_4_8 \|\| new_type == GGML_TYPE_Q4_0_8_8) {`
`16521`	`16518`	`new_type = GGML_TYPE_Q4_0;`
`16522`	`16519`	`}`