From 286e1fa56923a387a2015d9f31825c67c75c16fd Mon Sep 17 00:00:00 2001 From: ibrahimkhadraoui Date: Mon, 7 Jul 2025 14:36:51 +0400 Subject: [PATCH] fix rope_theta --- convert_hf_to_gguf.py | 2 +- src/llama-arch.cpp | 1 - src/llama-arch.h | 1 - src/llama-model.cpp | 1 - 4 files changed, 1 insertion(+), 4 deletions(-) diff --git a/convert_hf_to_gguf.py b/convert_hf_to_gguf.py index 5ef225d247..65017ab705 100755 --- a/convert_hf_to_gguf.py +++ b/convert_hf_to_gguf.py @@ -6675,7 +6675,7 @@ class FalconH1Model(Mamba2Model): self.gguf_writer.add_bool("falcon_h1.mamba_use_mlp", self.find_hparam(["mamba_use_mlp"], optional=True)) self.gguf_writer.add_bool("falcon_h1.mamba_norm_before_gate", self.find_hparam(["mamba_norm_before_gate"], optional=True)) self.gguf_writer.add_bool("falcon_h1.mamba_rms_norm", self.find_hparam(["mamba_rms_norm"], optional=True)) - self.gguf_writer.add_float64("falcon_h1.rope_theta", self.find_hparam(["rope_theta"], optional=True)) + self.gguf_writer.add_rope_freq_base(self.find_hparam(["rope_theta"])) ###### CONVERSION LOGIC ###### diff --git a/src/llama-arch.cpp b/src/llama-arch.cpp index ae6ea76fff..7fb81cfdc8 100644 --- a/src/llama-arch.cpp +++ b/src/llama-arch.cpp @@ -237,7 +237,6 @@ static const std::map LLM_KV_NAMES = { { LLM_KV_FALCON_H1_SSM_HAS_MUP, "%s.ssm.has_mup" }, { LLM_KV_FALCON_H1_MAMBA_NORM_BEFORE_GATE, "%s.mamba_norm_before_gate" }, { LLM_KV_FALCON_H1_MAMBA_RMS_NORM, "%s.mamba_rms_norm" }, - { LLM_KV_FALCON_H1_ROPE_THETA, "%s.rope_theta" }, { LLM_KV_FALCON_H1_KEY_MULTIPLIER, "%s.key_multiplier" }, { LLM_KV_FALCON_H1_LM_HEAD_MULTIPLIER, "%s.lm_head_multiplier" }, { LLM_KV_FALCON_H1_EMBEDDING_MULTIPLIER, "%s.embedding_multiplier" }, diff --git a/src/llama-arch.h b/src/llama-arch.h index c05cb85197..4ad1beb245 100644 --- a/src/llama-arch.h +++ b/src/llama-arch.h @@ -172,7 +172,6 @@ enum llm_kv { LLM_KV_FALCON_H1_SSM_HAS_MUP, LLM_KV_FALCON_H1_MAMBA_NORM_BEFORE_GATE, LLM_KV_FALCON_H1_MAMBA_RMS_NORM, - LLM_KV_FALCON_H1_ROPE_THETA, LLM_KV_FALCON_H1_KEY_MULTIPLIER, LLM_KV_FALCON_H1_LM_HEAD_MULTIPLIER, LLM_KV_FALCON_H1_EMBEDDING_MULTIPLIER, diff --git a/src/llama-model.cpp b/src/llama-model.cpp index 84188d16cf..fe7ba4f9ac 100644 --- a/src/llama-model.cpp +++ b/src/llama-model.cpp @@ -1577,7 +1577,6 @@ void llama_model::load_hparams(llama_model_loader & ml) { ml.get_key(LLM_KV_FALCON_H1_SSM_HAS_MUP, hparams.ssm_has_mup); ml.get_key(LLM_KV_FALCON_H1_MAMBA_NORM_BEFORE_GATE, hparams.mamba_norm_before_gate); ml.get_key(LLM_KV_FALCON_H1_MAMBA_RMS_NORM, hparams.mamba_rms_norm); - ml.get_key(LLM_KV_FALCON_H1_ROPE_THETA, hparams.rope_theta); ml.get_key(LLM_KV_FALCON_H1_KEY_MULTIPLIER, hparams.key_multiplier); ml.get_key(LLM_KV_FALCON_H1_LM_HEAD_MULTIPLIER, hparams.lm_head_multiplier); ml.get_key(LLM_KV_FALCON_H1_EMBEDDING_MULTIPLIER, hparams.embedding_multiplier);