From e14ef3eeda8d921ef3feaf3fae7bfe820ec69169 Mon Sep 17 00:00:00 2001
From: Wanis Elabbar <70503629+elabbarw@users.noreply.github.com>
Date: Tue, 23 Jul 2024 23:57:50 +0100
Subject: [PATCH 1/2] feat - add azure_ai llama v3.1 8B 70B and 405B

---
 model_prices_and_context_window.json | 27 +++++++++++++++++++++++++++
 1 file changed, 27 insertions(+)

diff --git a/model_prices_and_context_window.json b/model_prices_and_context_window.json
index f86ea8bd7..4c6dd8fdb 100644
--- a/model_prices_and_context_window.json
+++ b/model_prices_and_context_window.json
@@ -760,6 +760,33 @@
         "litellm_provider": "azure_ai",
         "mode": "chat"
     },
+    "azure_ai/Meta-Llama-31-8B-Instruct": {
+        "max_tokens": 128000,
+        "max_input_tokens": 128000,
+        "max_output_tokens": 128000,
+        "input_cost_per_token": 0.0000003,
+        "output_cost_per_token": 0.00000061,
+        "litellm_provider": "azure_ai",
+        "mode": "chat"
+    },
+    "azure_ai/Meta-Llama-31-70B-Instruct": {
+        "max_tokens": 128000,
+        "max_input_tokens": 128000,
+        "max_output_tokens": 128000,
+        "input_cost_per_token": 0.00000268,
+        "output_cost_per_token": 0.00000354,
+        "litellm_provider": "azure_ai",
+        "mode": "chat"
+    },
+    "azure_ai/Meta-Llama-31-405B-Instruct": {
+        "max_tokens": 128000,
+        "max_input_tokens": 128000,
+        "max_output_tokens": 128000,
+        "input_cost_per_token": 0.00000533,
+        "output_cost_per_token": 0.000016,
+        "litellm_provider": "azure_ai",
+        "mode": "chat"
+    },
     "babbage-002": {
         "max_tokens": 16384,
         "max_input_tokens": 16384,

From 77cf1fd600b5213422e807de277e6599a03a45dd Mon Sep 17 00:00:00 2001
From: Wanis Elabbar <70503629+elabbarw@users.noreply.github.com>
Date: Wed, 24 Jul 2024 16:50:07 +0100
Subject: [PATCH 2/2] update azure_ai llamav31 prices with sources

---
 model_prices_and_context_window.json | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)

diff --git a/model_prices_and_context_window.json b/model_prices_and_context_window.json
index e8ca6f74d..90a6b1283 100644
--- a/model_prices_and_context_window.json
+++ b/model_prices_and_context_window.json
@@ -767,7 +767,8 @@
         "input_cost_per_token": 0.0000003,
         "output_cost_per_token": 0.00000061,
         "litellm_provider": "azure_ai",
-        "mode": "chat"
+        "mode": "chat",
+        "source":"https://azuremarketplace.microsoft.com/en-us/marketplace/apps/metagenai.meta-llama-3-1-8b-instruct-offer?tab=PlansAndPrice"
     },
     "azure_ai/Meta-Llama-31-70B-Instruct": {
         "max_tokens": 128000,
@@ -776,7 +777,8 @@
         "input_cost_per_token": 0.00000268,
         "output_cost_per_token": 0.00000354,
         "litellm_provider": "azure_ai",
-        "mode": "chat"
+        "mode": "chat",
+        "source":"https://azuremarketplace.microsoft.com/en-us/marketplace/apps/metagenai.meta-llama-3-1-70b-instruct-offer?tab=PlansAndPrice"
     },
     "azure_ai/Meta-Llama-31-405B-Instruct": {
         "max_tokens": 128000,
@@ -785,7 +787,8 @@
         "input_cost_per_token": 0.00000533,
         "output_cost_per_token": 0.000016,
         "litellm_provider": "azure_ai",
-        "mode": "chat"
+        "mode": "chat",
+        "source":"https://azuremarketplace.microsoft.com/en-us/marketplace/apps/metagenai.meta-llama-3-1-405b-instruct-offer?tab=PlansAndPrice"
     },
     "babbage-002": {
         "max_tokens": 16384,