From: Faisal Zaghloul <redacted>
Date: Thu, 7 Nov 2024 16:46:12 +0000 (-0500)
Subject: Remove identical wte/etw logic for jais (#10203)
X-Git-Tag: upstream/0.0.4488~444
X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=60e17ce23c2740369af6304113a2dfa0454eaf26;p=pkg%2Fggml%2Fsources%2Fllama.cpp

Remove identical wte/etw logic for jais (#10203)
---

diff --git a/convert_hf_to_gguf.py b/convert_hf_to_gguf.py
index 76ee6cef..39afa5ef 100755
--- a/convert_hf_to_gguf.py
+++ b/convert_hf_to_gguf.py
@@ -3748,10 +3748,7 @@ class JaisModel(Model):
 
         # Embeddings scale
         self.embeddings_scale = 1.0
-        # note: For some JAIS flavors, output is tied to (same as) wte in original model
-        self.output_is_wte = False
         if 'mup_embeddings_scale' in self.hparams:
-            self.output_is_wte = True   # Hack (?)
             self.embeddings_scale = self.hparams['mup_embeddings_scale']
         elif 'embeddings_scale' in self.hparams:
             self.embeddings_scale = self.hparams['embeddings_scale']
@@ -3808,10 +3805,7 @@ class JaisModel(Model):
 
         if new_name == self.format_tensor_name(gguf.MODEL_TENSOR.TOKEN_EMBD):
             tensors.append((new_name, data_torch * self.embeddings_scale))
-            if self.output_is_wte:
-                tensors.append((self.format_tensor_name(gguf.MODEL_TENSOR.OUTPUT), data_torch * self.width_scale))
         elif new_name == self.format_tensor_name(gguf.MODEL_TENSOR.OUTPUT):
-            assert not self.output_is_wte
             tensors.append((new_name, data_torch * self.width_scale))
         else:
             tensors.append((new_name, data_torch))