Add additional Falcon convert requirements

Author: countzero

.gitignore

@@ -2,5 +2,5 @@
 # .gitignore
 #
 
-# Ignore Sublime Text files.
 *.sublime-workspace
+*.log

examples/server_llama_2_13b_chat.ps1

@@ -0,0 +1,7 @@
+Start-Process "http://127.0.0.1:8080"
+
+../vendor/llama.cpp/build/bin/Release/server `
+    --model "../vendor/llama.cpp/models/Llama-2-13b-chat-hf/model-quantized-q4_K_M.gguf" `
+    --ctx-size 4096 `
+    --threads 16 `
+    --n-gpu-layers 20

examples/server_llama_2_7b_chat.ps1

@@ -0,0 +1,7 @@
+Start-Process "http://127.0.0.1:8080"
+
+../vendor/llama.cpp/build/bin/Release/server `
+    --model "../vendor/llama.cpp/models/Llama-2-7b-chat-hf/model-quantized-q4_K_M.gguf" `
+    --ctx-size 4096 `
+    --threads 16 `
+    --n-gpu-layers 35

examples/server_phind_codellama_34b_v2.ps1

@@ -0,0 +1,7 @@
+Start-Process "http://127.0.0.1:8080"
+
+../vendor/llama.cpp/build/bin/Release/server `
+    --model "../vendor/llama.cpp/models/Phind-CodeLlama-34B-v2/model-quantized-q4_K_M.gguf" `
+    --ctx-size 4096 `
+    --threads 16 `
+    --n-gpu-layers 10

examples/server_upstage_llama_2_70b_instruct_v2.ps1

@@ -0,0 +1,7 @@
+Start-Process "http://127.0.0.1:8080"
+
+../vendor/llama.cpp/build/bin/Release/server `
+    --model "../vendor/llama.cpp/models/Llama-2-70b-instruct-v2/model-quantized-q4_K_M.gguf" `
+    --ctx-size 4096 `
+    --threads 16 `
+    --n-gpu-layers 0

rebuild_llama.cpp.ps1

@@ -147,11 +147,14 @@ Set-Location -Path "../"
 
 conda activate llama.cpp
 
-# We are making sure to always use the latest version.
+# We are making sure to always use the latest version of the "gguf" package.
 pip install --ignore-installed -r ./requirements.txt
 
 Set-Location -Path "../../"
 
+# We also need to install additional packages to support falcon conversion.
+pip install -r ./requirements.txt
+
 $stopwatch.Stop()
 $durationInSeconds = [Math]::Floor([Decimal]($stopwatch.Elapsed.TotalSeconds))
 

requirements.txt

@@ -0,0 +1,6 @@
+# We are using a specific version of the "torch"
+# package which supports a specific CUDA version.
+--extra-index-url https://download.pytorch.org/whl/nightly/cu121
+torch==2.1.0.dev20230905+cu121
+
+transformers==4.33.1