feat: add codestral (#253)

Signed-off-by: Sertac Ozercan <sozercan@gmail.com>

.github/workflows/patch-models.yaml

@@ -25,6 +25,7 @@ jobs:
               - ghcr.io/sozercan/mixtral:8x7b
               - ghcr.io/sozercan/phi3:3.8b
               - ghcr.io/sozercan/gemma1.1:2b
+              - ghcr.io/sozercan/codestral0.1:22b
         steps:
         - uses: jlumbroso/free-disk-space@54081f138730dfa15788a46383842cd2f914a1be # v1.3.1
           with:

.github/workflows/update-models.yaml

@@ -25,6 +25,7 @@ jobs:
          - llama-3-8b-instruct
          - phi-3-3.8b
          - gemma-1.1-2b-instruct
+         - codestral-0.1-22b
     runs-on: ubuntu-latest
     timeout-minutes: 360
     steps:

models/codestral-0.1-22b.yaml

@@ -0,0 +1,31 @@
+#syntax=ghcr.io/sozercan/aikit:latest
+apiVersion: v1alpha1
+debug: true
+runtime: cuda
+models:
+  - name: codestral-0.1-22b
+    source: https://huggingface.co/lmstudio-community/Codestral-22B-v0.1-GGUF/resolve/main/Codestral-22B-v0.1-Q4_K_M.gguf
+    sha256: defc9e0a1bb42857558d43df4e7f0f3d0a29d06a953e498e967d763f45d10431
+    promptTemplates:
+      - name: instruct
+        template: |
+          [INST]{{ if .SystemPrompt }}{{ .SystemPrompt }}{{ end }} {{ .Input }}[/INST]
+config: |
+  - name: codestral-0.1-22b
+    backend: llama
+    parameters:
+      model: Codestral-22B-v0.1-Q4_K_M.gguf
+    context_size: 8192
+    template:
+      chat: instruct
+      completion: instruct
+    stopwords:
+     - \"[INST]\"
+     - \"[/INST]\"
+     - \"[PREFIX]\"
+     - \"[MIDDLE]\"
+     - \"[SUFFIX]\"
+    gpu_layers: 50
+    f16: true
+    batch: 512
+    mmap: true