Add Qwen2-72B model (hosted by DeepInfra), remove Dolphin 8x7B (#58)

Dolphin is now often unavailable on DeepInfra, so it has been retired. Also increase max output tokens to 100,000 (maximum).
XInTheDark · Jun 30, 2024 · 7fc58f0 · 7fc58f0
1 parent a67f496
commit 7fc58f0
Show file tree

Hide file tree

Showing 4 changed files with 9 additions and 9 deletions.
diff --git a/README.md b/README.md
@@ -59,11 +59,11 @@ updating manually also allows you to fetch and view the latest changes to the so
 | GPT                | gpt-3.5-turbo (default) | ✅         | ![Active][active-badge]   | Fast      | 7.5/10, the most reliable and decently performing model but there are some stronger models.   |
 | GPT                | gpt-4                   | ❌         | ![Active][active-badge]   | Medium    | 6/10, no streaming support but otherwise a great model.                                       |
 | Bing               | gpt-4                   | ✅         | ![Unknown][unknown-badge] | Slow      | 6/10, generation speed is likely quite slow but comes with built-in web search ability.       |
-| DeepInfra          | WizardLM-2-8x22B        | ✅         | ![Active][active-badge]   | Fast      | 8/10, very capable model for all purposes.                                                    |
-| DeepInfra          | meta-llama-3-8b         | ✅         | ![Active][active-badge]   | Very fast | 7/10                                                                                          |
-| DeepInfra          | meta-llama-3-70b        | ✅         | ![Active][active-badge]   | Medium    | 7.5/10                                                                                        |
+| DeepInfra          | WizardLM-2-8x22B        | ✅         | ![Active][active-badge]   | Medium    | 7.5/10, capable model for general use.                                                        |
+| DeepInfra          | meta-llama-3-8b         | ✅         | ![Active][active-badge]   | Very fast | 6.5/10                                                                                        |
+| DeepInfra          | meta-llama-3-70b        | ✅         | ![Active][active-badge]   | Medium    | 7/10                                                                                          |
 | DeepInfra          | Mixtral-8x22B           | ✅         | ![Active][active-badge]   | Fast      | 7/10                                                                                          |
-| DeepInfra          | Dolphin-2.6-8x7B        | ✅         | ![Active][active-badge]   | Fast      | 5/10                                                                                          |
+| DeepInfra          | Qwen2-72B               | ✅         | ![Active][active-badge]   | Fast      | 7.5/10                                                                                        |
 | Blackbox           | custom model            | ✅         | ![Active][active-badge]   | Very fast | 6.5/10, very fast generation with built-in web search ability, but is optimized for coding.   |
 | Ecosia             | gpt-3.5-turbo           | ✅         | ![Active][active-badge]   | Very fast | 7.5/10, near instant responses with a recent model.                                           |
 | Replicate          | meta-llama-3-8b         | ✅         | ![Active][active-badge]   | Fast      | ?/10                                                                                          |

diff --git a/package.json b/package.json
@@ -222,8 +222,8 @@
           "value": "DeepInfraMixtral_8x22B"
         },
         {
-          "title": "DeepInfra (Dolphin-2.6-8x7B)",
-          "value": "DeepInfraDolphin26_8x7B"
+          "title": "DeepInfra (Qwen2-72B)",
+          "value": "DeepInfraQwen2_72B"
         },
         {
           "title": "Blackbox (custom-model)",

diff --git a/src/api/Providers/deepinfra.jsx b/src/api/Providers/deepinfra.jsx
@@ -31,7 +31,7 @@ export const getDeepInfraResponse = async function* (chat, options, max_retries
     model: model,
     messages: chat,
     temperature: options.temperature ?? 0.7,
-    max_tokens: model.includes("Meta-Llama-3") ? 1028 : null,
+    max_tokens: 100000,
     stream: true,
     headers: headers,
   };

diff --git a/src/api/providers.jsx b/src/api/providers.jsx
@@ -48,7 +48,7 @@ export const providers_info = {
   DeepInfraLlama3_8B: { provider: DeepInfraProvider, model: "meta-llama/Meta-Llama-3-8B-Instruct", stream: true },
   DeepInfraLlama3_70B: { provider: DeepInfraProvider, model: "meta-llama/Meta-Llama-3-70B-Instruct", stream: true },
   DeepInfraMixtral_8x22B: { provider: DeepInfraProvider, model: "mistralai/Mixtral-8x22B-Instruct-v0.1", stream: true },
-  DeepInfraDolphin26_8x7B: { provider: DeepInfraProvider, model: "cognitivecomputations/dolphin-2.6-mixtral-8x7b", stream: true, },
+  DeepInfraQwen2_72B: { provider: DeepInfraProvider, model: "Qwen/Qwen2-72B-Instruct", stream: true },
   Blackbox: { provider: BlackboxProvider, model: "", stream: true },
   Ecosia: { provider: EcosiaProvider, model: "gpt-3.5-turbo-0125", stream: true },
   ReplicateLlama3_8B: { provider: ReplicateProvider, model: "meta/meta-llama-3-8b-instruct", stream: true },
@@ -67,7 +67,7 @@ export const chat_providers = [
   ["DeepInfra (meta-llama-3-8b)", "DeepInfraLlama3_8B"],
   ["DeepInfra (meta-llama-3-70b)", "DeepInfraLlama3_70B"],
   ["DeepInfra (Mixtral-8x22B)", "DeepInfraMixtral_8x22B"],
-  ["DeepInfra (Dolphin-2.6-8x7B)", "DeepInfraDolphin26_8x7B"],
+  ["DeepInfra (Qwen2-72B)", "DeepInfraQwen2_72B"],
   ["Blackbox (custom-model)", "Blackbox"],
   ["Ecosia (gpt-3.5-turbo)", "Ecosia"],
   ["Replicate (meta-llama-3-8b)", "ReplicateLlama3_8B"],