Rewrite to save memory

Radiohead_scraper.ipynb

@@ -0,0 +1,116 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from lyricsgenius import Genius\n",
+    "import os\n",
+    "import time\n",
+    "\n",
+    "# Replace with your Genius API key\n",
+    "GENIUS_API_KEY = os.environ.get(\"GENIUS_TOKEN\")\n",
+    "\n",
+    "#https://lyricsgenius.readthedocs.io/en/master/usage.html\n",
+    "genius = Genius(GENIUS_API_KEY)   \n",
+    "genius.remove_section_headers = True  # Removes section headers from lyrics\n",
+    "genius.excluded_terms = [\"(Remix)\", \"(Live)\", \"(Acoustic)\", \"(Demo)\"]  # Exclude these terms from song titles\n",
+    "#albums_list = genius.artist_albums(1421, per_page=None, page=None)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "album_list = [\"Kid A\", \"OK Computer\", \"In Rainbows\", \"A Moon Shaped Pool\", \"Hail to the Thief\", \"The King of Limbs\",'The Bends',]\n",
+    "album_list_2= [\"Pablo Honey\", \"Amnesiac\",]\n",
+    "               # \"A Brief History of Love\", \"The Eraser\", \"Tomorrow's Modern Boxes\",\n",
+    "              #\"Anima\", \"The Smile\", \"Kid A Mnesia\", \"I Might Be Wrong\", \"Live in Berlin\", \"Com Lag (2plus2isfive)\"]"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 3,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "genius.sleep_time=3\n",
+    "genius.timeout = 15  # Set timeout to 15 seconds to avoid rate limiting"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 4,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Searching for \"Pablo Honey\" by Radiohead...\n",
+      "Wrote lyrics for Pablo Honey.\n",
+      "Searching for \"Amnesiac\" by Radiohead...\n",
+      "Wrote lyrics for Amnesiac.\n",
+      "Searching for \"A Brief History of Love\" by Radiohead...\n",
+      "No results found for: 'A Brief History of Love Radiohead'\n"
+     ]
+    },
+    {
+     "ename": "AttributeError",
+     "evalue": "'NoneType' object has no attribute 'to_dict'",
+     "output_type": "error",
+     "traceback": [
+      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
+      "\u001b[0;31mAttributeError\u001b[0m                            Traceback (most recent call last)",
+      "Cell \u001b[0;32mIn[4], line 3\u001b[0m\n\u001b[1;32m      1\u001b[0m \u001b[38;5;28;01mfor\u001b[39;00m album_name \u001b[38;5;129;01min\u001b[39;00m album_list_2:\n\u001b[1;32m      2\u001b[0m     album \u001b[38;5;241m=\u001b[39m genius\u001b[38;5;241m.\u001b[39msearch_album(album_name, \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mRadiohead\u001b[39m\u001b[38;5;124m\"\u001b[39m,)  \u001b[38;5;66;03m# Search for the album by title and artist\u001b[39;00m\n\u001b[0;32m----> 3\u001b[0m     album_dict \u001b[38;5;241m=\u001b[39m \u001b[43malbum\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mto_dict\u001b[49m()  \u001b[38;5;66;03m# Convert the album object to a dictionary\u001b[39;00m\n\u001b[1;32m      4\u001b[0m     album_length \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mlen\u001b[39m(album_dict[\u001b[38;5;124m'\u001b[39m\u001b[38;5;124mtracks\u001b[39m\u001b[38;5;124m'\u001b[39m])\n\u001b[1;32m      5\u001b[0m     \u001b[38;5;28;01mwith\u001b[39;00m \u001b[38;5;28mopen\u001b[39m(\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mradiohead_lyrics.txt\u001b[39m\u001b[38;5;124m\"\u001b[39m, \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124ma\u001b[39m\u001b[38;5;124m\"\u001b[39m, encoding\u001b[38;5;241m=\u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mutf-8\u001b[39m\u001b[38;5;124m\"\u001b[39m) \u001b[38;5;28;01mas\u001b[39;00m file:\n",
+      "\u001b[0;31mAttributeError\u001b[0m: 'NoneType' object has no attribute 'to_dict'"
+     ]
+    }
+   ],
+   "source": [
+    "for album_name in album_list_2:\n",
+    "    album = genius.search_album(album_name, \"Radiohead\",)  # Search for the album by title and artist\n",
+    "    album_dict = album.to_dict()  # Convert the album object to a dictionary\n",
+    "    album_length = len(album_dict['tracks'])\n",
+    "    with open(\"radiohead_lyrics.txt\", \"a\", encoding=\"utf-8\") as file:\n",
+    "        for song in range(album_length):\n",
+    "            lyrics = album_dict['tracks'][song]['song']['lyrics'].split('Lyrics\\n')[-1]\n",
+    "            file.write(lyrics + \"\\n\\n\" + \"=\" * 50 + \"\\n\\n\")  # Lyrics + divider\n",
+    "    print(f\"Wrote lyrics for {album_name}.\")\n",
+    "    \n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "3.10.16",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.10.16"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}

app.py

@@ -99,12 +99,10 @@ def artist_response(gemma_response, artist):
         lyric_list = all_phrases_grateful_dead
     if artist == "Google Gemma":
         return gemma_response
-
-    encoded_gemma = encoder_model.encode(gemma_response)
-    similarity_result = encoder_model.similarity(
-            encoded_gemma,
-            artist_embeddings,
-        )
+    encoder = get_encoder()
+    encoded_gemma = encoder.encode(gemma_response, precision="int8")
+    #encoded_gemma = encoder_model.encode(gemma_response)
+    similarity_result = cosine_similarity_int8(encoded_gemma, artist_embeddings)
     result_max_index = np.argmax(similarity_result)
     lyric_response = lyric_list[result_max_index]
 
@@ -146,7 +144,7 @@ def chat_with_musician(user_input, history, artist):
             try:
                 response = client.chat_completion(
                     messages=messages,
-                    model="google/gemma-2-2b-it",
+                    #model="google/gemma-2-2b-it",
                     max_tokens=256,
                     temperature=0.75,
                     #top_p=0.9
@@ -162,36 +160,39 @@ def chat_with_musician(user_input, history, artist):
     artist_history[:] = artist_history[-10:]  # Keep only last 10 entries
     return lyric_response
 
+def cosine_similarity_int8(query, embeddings):
+    # query: (d,)
+    # embeddings: (n, d)
+    query = query.astype(np.int32)
+    embeddings = embeddings.astype(np.int32)
+
+    dots = embeddings @ query
+    query_norm = np.linalg.norm(query)
+    emb_norms = np.linalg.norm(embeddings, axis=1)
+
+    return dots / (emb_norms * query_norm + 1e-8)
+
 HF_API_KEY = os.environ["HF_API_KEY"]
 
-radiohead_lyrics = load_lyrics("radiohead_lyrics.txt")
-kendrick_lyrics = load_lyrics("kendrick_lamar_lyrics.txt")
-grateful_dead_lyrics = load_lyrics('grateful_dead_lyrics.txt')
 
-all_phrases_radiohead = generate_cumulative_phrases(songs_from_text(radiohead_lyrics))
-all_phrases_kendrick = generate_cumulative_phrases(songs_from_text(kendrick_lyrics))
-all_phrases_grateful_dead = generate_cumulative_phrases(songs_from_text(grateful_dead_lyrics))
 
-encoder_model = SentenceTransformer('all-MiniLM-L6-v2',
+_encoder_model = None
+
+def get_encoder():
+    global _encoder_model
+    if _encoder_model is None:
+        _encoder_model = SentenceTransformer('all-MiniLM-L6-v2',
                                     #'sentence-transformers/all-MiniLM-L6-v2', 
-                                    backend='openvino',
-                                    model_kwargs={"file_name": "openvino/openvino_model_qint8_quantized.xml"},
+                                    #backend='openvino',
+                                    #model_kwargs={"file_name": "openvino/openvino_model_qint8_quantized.xml"},
                                     #to increase speed:
                                     #similarity_function=SimilarityFunction.DOT_PRODUCT,
                                     )
-radiohead_embeddings = encoder_model.encode(all_phrases_radiohead,
-                        precision="int8",
-                        show_progress_bar=True,)
-radiohead_embeddings = radiohead_embeddings.astype(np.float32)
-kendrick_embeddings = encoder_model.encode(all_phrases_kendrick,
-                                            precision="int8",
-                                           show_progress_bar=True,)
-kendrick_embeddings = kendrick_embeddings.astype(np.float32)
-grateful_dead_embeddings = encoder_model.encode(all_phrases_grateful_dead,
-                                               precision='int8',
-                                               show_progress_bar=True,)
-grateful_dead_embeddings = grateful_dead_embeddings.astype(np.float32)
+    return _encoder_model
 
+radiohead_embeddings = np.load("radiohead_embeddings.npy", mmap_mode="r")
+kendrick_embeddings = np.load("kendrick_embeddings.npy", mmap_mode="r")
+grateful_dead_embeddings = np.load("grateful_dead_embeddings.npy", mmap_mode="r")
 
 
 size = 350 #256

grateful_dead_embeddings.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0134ba4610693e490146b62ac478e93ad351d1e07b0ef09560b229d2cb4aad10
+size 18111104

kendrick_embeddings.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a333db77a592551e63fcdfd92d4f0529b60cdb4e99c3936220068a876ce28593
+size 9623168

radiohead_embeddings.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:da9a1b03efd746fa27fb30ea9e401c46fe87bc5b592bb3ebbdacd1b4b7843b0f
+size 6394496