From dcd80a126c6232c3a9ed82b5c877200999308ebc Mon Sep 17 00:00:00 2001
From: saloui <saifeddine.aloui@cea.fr>
Date: Tue, 12 Dec 2023 16:49:51 +0100
Subject: [PATCH] enhanced generation

---
 lollms/media.py       |  3 ++-
 lollms/personality.py |  2 +-
 lollms/utilities.py   | 24 ++++++++++++++----------
 3 files changed, 17 insertions(+), 12 deletions(-)

diff --git a/lollms/media.py b/lollms/media.py
index 954d732..d442f90 100644
--- a/lollms/media.py
+++ b/lollms/media.py
@@ -68,7 +68,7 @@ class AudioRecorder:
     def start_recording(self):
         if self.whisper_model is None:
             self.lollmsCom.info("Loading whisper model")
-            self.whisper_model=whisper.load_model("base")
+            self.whisper_model=whisper.load_model("base.en")
         try:
             self.is_recording = True
             self.audio_stream = pyaudio.PyAudio().open(
@@ -123,6 +123,7 @@ class AudioRecorder:
                         # Convert to float
 
                         audio_data = self.audio_frames.astype(np.float32)
+                        
                         # Transcribe the audio using the whisper model
                         text = self.whisper_model.transcribe(audio_data[non_silent_start:non_silent_end])
 
diff --git a/lollms/personality.py b/lollms/personality.py
index c550561..84aab03 100644
--- a/lollms/personality.py
+++ b/lollms/personality.py
@@ -715,7 +715,7 @@ Date: {{date}}
                         idx = pth.index("uploads")
                         pth = "/".join(pth[idx:])
                         self.new_message("",MSG_TYPE.MSG_TYPE_FULL)
-                        output = f'<img src="{pth}" width="300">\n\n'
+                        output = f'<img src="{pth}" width="800">\n\n'
                         self.full(output)
 
                     if self.model.binding_type not in [BindingType.TEXT_IMAGE, BindingType.TEXT_IMAGE_VIDEO]:
diff --git a/lollms/utilities.py b/lollms/utilities.py
index 87b6028..b09c7b9 100644
--- a/lollms/utilities.py
+++ b/lollms/utilities.py
@@ -576,18 +576,22 @@ class PromptReshaper:
         def fill_template(template, data):
             for key, value in data.items():
                 placeholder = "{{" + key + "}}"
-                template = template.replace(placeholder, value)
+                n_text_tokens = len(tokenize(template))
+                if key in place_holders_to_sacrifice:
+                    n_remaining = max_nb_tokens - n_text_tokens
+                    t_value = tokenize(value)
+                    n_value = len(t_value)
+                    if n_value<n_remaining:
+                        template = template.replace(placeholder, value)
+                    else:
+                        value = detokenize(t_value[-n_remaining:])
+                        template = template.replace(placeholder, value)
+                        
+                else:
+                    template = template.replace(placeholder, value)
             return template
         
-        if max_nb_tokens-all_count>0 or len(place_holders_to_sacrifice)==0:
-            return fill_template(self.template, placeholders)
-        else:
-            to_remove = -int((max_nb_tokens - all_count)/len(place_holders_to_sacrifice))
-            for placeholder, text in placeholders.items():
-                if placeholder in place_holders_to_sacrifice:
-                    text_tokens = tokenize(text)[to_remove:]
-                    placeholders[placeholder]=detokenize(text_tokens)
-            return fill_template(self.template, placeholders)
+        return fill_template(self.template, placeholders)