diff options
Diffstat (limited to 'src/llama.cpp')
| -rw-r--r-- | src/llama.cpp | 13 |
1 files changed, 9 insertions, 4 deletions
diff --git a/src/llama.cpp b/src/llama.cpp index 3f9a211c..4e7e4a6c 100644 --- a/src/llama.cpp +++ b/src/llama.cpp @@ -3761,8 +3761,9 @@ struct llama_model_loader { case GGML_TYPE_IQ2_BN: ftype = LLAMA_FTYPE_MOSTLY_IQ2_BN; break; case GGML_TYPE_IQ4_NL: ftype = LLAMA_FTYPE_MOSTLY_IQ4_NL; break; case GGML_TYPE_IQ4_XS: ftype = LLAMA_FTYPE_MOSTLY_IQ4_XS; break; - case GGML_TYPE_IQ4_K: ftype = LLAMA_FTYPE_MOSTLY_IQ4_K; break; case GGML_TYPE_IQ2_K: ftype = LLAMA_FTYPE_MOSTLY_IQ2_K; break; + case GGML_TYPE_IQ4_K: ftype = LLAMA_FTYPE_MOSTLY_IQ4_K; break; + case GGML_TYPE_IQ5_K: ftype = LLAMA_FTYPE_MOSTLY_IQ5_K; break; case GGML_TYPE_IQ3_S: ftype = LLAMA_FTYPE_MOSTLY_IQ3_S; break; case GGML_TYPE_Q4_0_4_4: ftype = LLAMA_FTYPE_MOSTLY_Q4_0_4_4; break; case GGML_TYPE_Q4_0_4_8: ftype = LLAMA_FTYPE_MOSTLY_Q4_0_4_8; break; @@ -4458,8 +4459,9 @@ static std::string llama_model_ftype_name(llama_ftype ftype) { case LLAMA_FTYPE_MOSTLY_IQ1_M: return "IQ1_M - 1.75 bpw"; case LLAMA_FTYPE_MOSTLY_IQ4_NL: return "IQ4_NL - 4.5 bpw"; case LLAMA_FTYPE_MOSTLY_IQ4_XS: return "IQ4_XS - 4.25 bpw"; - case LLAMA_FTYPE_MOSTLY_IQ4_K: return "IQ4_K - 4.5 bpw"; case LLAMA_FTYPE_MOSTLY_IQ2_K: return "IQ2_K - 2.375 bpw"; + case LLAMA_FTYPE_MOSTLY_IQ4_K: return "IQ4_K - 4.5 bpw"; + case LLAMA_FTYPE_MOSTLY_IQ5_K: return "IQ5_K - 5.5 bpw"; 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"; @@ -15635,7 +15637,8 @@ static ggml_type llama_tensor_get_type(quantize_state_internal & qs, ggml_type n new_type == GGML_TYPE_Q5_K || new_type == GGML_TYPE_Q6_K || new_type == GGML_TYPE_IQ4_XS || new_type == GGML_TYPE_IQ2_XS || new_type == GGML_TYPE_IQ2_XXS || new_type == GGML_TYPE_IQ2_S || new_type == GGML_TYPE_IQ3_XXS || new_type == GGML_TYPE_IQ1_S || new_type == GGML_TYPE_IQ3_S || - new_type == GGML_TYPE_IQ1_M || new_type == GGML_TYPE_IQ4_K || new_type == GGML_TYPE_IQ2_K) { + new_type == GGML_TYPE_IQ1_M || new_type == GGML_TYPE_IQ4_K || new_type == GGML_TYPE_IQ2_K || + new_type == GGML_TYPE_IQ5_K) { int nx = tensor->ne[0]; int ny = tensor->ne[1]; if (nx % QK_K != 0) { @@ -15666,6 +15669,7 @@ static ggml_type llama_tensor_get_type(quantize_state_internal & qs, ggml_type n case GGML_TYPE_IQ4_XS: new_type = GGML_TYPE_IQ4_NL; break; case GGML_TYPE_IQ4_K: case GGML_TYPE_Q4_K: new_type = GGML_TYPE_Q5_0; break; + case GGML_TYPE_IQ5_K: case GGML_TYPE_Q5_K: new_type = GGML_TYPE_Q5_1; break; case GGML_TYPE_Q6_K: new_type = GGML_TYPE_Q8_0; break; default: throw std::runtime_error("\nUnsupported tensor size encountered\n"); @@ -15768,8 +15772,9 @@ static void llama_model_quantize_internal(const std::string & fname_inp, const s case LLAMA_FTYPE_MOSTLY_IQ2_BN: default_type = GGML_TYPE_IQ2_BN; break; case LLAMA_FTYPE_MOSTLY_IQ4_NL: default_type = GGML_TYPE_IQ4_NL; break; case LLAMA_FTYPE_MOSTLY_IQ4_XS: default_type = GGML_TYPE_IQ4_XS; break; - case LLAMA_FTYPE_MOSTLY_IQ4_K: default_type = GGML_TYPE_IQ4_K; break; case LLAMA_FTYPE_MOSTLY_IQ2_K: default_type = GGML_TYPE_IQ2_K; break; + case LLAMA_FTYPE_MOSTLY_IQ4_K: default_type = GGML_TYPE_IQ4_K; break; + case LLAMA_FTYPE_MOSTLY_IQ5_K: default_type = GGML_TYPE_IQ5_K; break; case LLAMA_FTYPE_MOSTLY_IQ3_S: default_type = GGML_TYPE_IQ3_S; break; case LLAMA_FTYPE_MOSTLY_IQ3_M: default_type = GGML_TYPE_IQ3_S; break; case LLAMA_FTYPE_MOSTLY_Q4_0_4_4: default_type = GGML_TYPE_Q4_0_4_4; break; |