Remove Qwen2 model (#18444)

Removed deprecated Qwen2 7B Instruct model from zed.dev provider (staff
only).

Release Notes:

- N/A

crates/collab/k8s/collab.template.yml

@@ -149,18 +149,6 @@ spec:
                 secretKeyRef:
                   name: google-ai
                   key: api_key
-            - name: RUNPOD_API_KEY
-              valueFrom:
-                secretKeyRef:
-                  name: runpod
-                  key: api_key
-                  optional: true
-            - name: RUNPOD_API_SUMMARY_URL
-              valueFrom:
-                secretKeyRef:
-                  name: runpod
-                  key: summary
-                  optional: true
             - name: BLOB_STORE_ACCESS_KEY
               valueFrom:
                 secretKeyRef:

crates/collab/src/lib.rs

@@ -170,8 +170,6 @@ pub struct Config {
     pub anthropic_api_key: Option<Arc<str>>,
     pub anthropic_staff_api_key: Option<Arc<str>>,
     pub llm_closed_beta_model_name: Option<Arc<str>>,
-    pub runpod_api_key: Option<Arc<str>>,
-    pub runpod_api_summary_url: Option<Arc<str>>,
     pub zed_client_checksum_seed: Option<String>,
     pub slack_panics_webhook: Option<String>,
     pub auto_join_channel_id: Option<ChannelId>,
@@ -235,8 +233,6 @@ impl Config {
             stripe_api_key: None,
             stripe_price_id: None,
             supermaven_admin_api_key: None,
-            runpod_api_key: None,
-            runpod_api_summary_url: None,
             user_backfiller_github_access_token: None,
         }
     }

crates/collab/src/llm.rs

@@ -400,42 +400,6 @@ async fn perform_completion(
                 })
                 .boxed()
         }
-        LanguageModelProvider::Zed => {
-            let api_key = state
-                .config
-                .runpod_api_key
-                .as_ref()
-                .context("no Qwen2-7B API key configured on the server")?;
-            let api_url = state
-                .config
-                .runpod_api_summary_url
-                .as_ref()
-                .context("no Qwen2-7B URL configured on the server")?;
-            let chunks = open_ai::stream_completion(
-                &state.http_client,
-                api_url,
-                api_key,
-                serde_json::from_str(params.provider_request.get())?,
-                None,
-            )
-            .await?;
-
-            chunks
-                .map(|event| {
-                    event.map(|chunk| {
-                        let input_tokens =
-                            chunk.usage.as_ref().map_or(0, |u| u.prompt_tokens) as usize;
-                        let output_tokens =
-                            chunk.usage.as_ref().map_or(0, |u| u.completion_tokens) as usize;
-                        (
-                            serde_json::to_vec(&chunk).unwrap(),
-                            input_tokens,
-                            output_tokens,
-                        )
-                    })
-                })
-                .boxed()
-        }
     };
 
     Ok(Response::new(Body::wrap_stream(TokenCountingStream {

crates/collab/src/llm/authorization.rs

@@ -77,7 +77,6 @@ fn authorize_access_for_country(
         LanguageModelProvider::Anthropic => anthropic::is_supported_country(country_code),
         LanguageModelProvider::OpenAi => open_ai::is_supported_country(country_code),
         LanguageModelProvider::Google => google_ai::is_supported_country(country_code),
-        LanguageModelProvider::Zed => true,
     };
     if !is_country_supported_by_provider {
         Err(Error::http(
@@ -213,7 +212,6 @@ mod tests {
             (LanguageModelProvider::Anthropic, "T1"), // Tor
             (LanguageModelProvider::OpenAi, "T1"),    // Tor
             (LanguageModelProvider::Google, "T1"),    // Tor
-            (LanguageModelProvider::Zed, "T1"),       // Tor
         ];
 
         for (provider, country_code) in cases {

crates/collab/src/llm/db/seed.rs

@@ -40,15 +40,6 @@ pub async fn seed_database(_config: &Config, db: &mut LlmDatabase, _force: bool)
             price_per_million_input_tokens: 25,   // $0.25/MTok
             price_per_million_output_tokens: 125, // $1.25/MTok
         },
-        ModelParams {
-            provider: LanguageModelProvider::Zed,
-            name: "Qwen/Qwen2-7B-Instruct".into(),
-            max_requests_per_minute: 5,
-            max_tokens_per_minute: 25_000, // These are arbitrary limits we've set to cap costs; we control this number
-            max_tokens_per_day: 300_000,
-            price_per_million_input_tokens: 25,
-            price_per_million_output_tokens: 125,
-        },
     ])
     .await
 }

crates/collab/src/llm/db/tests/provider_tests.rs

@@ -26,7 +26,6 @@ async fn test_initialize_providers(db: &mut LlmDatabase) {
             LanguageModelProvider::Anthropic,
             LanguageModelProvider::Google,
             LanguageModelProvider::OpenAi,
-            LanguageModelProvider::Zed
         ]
     )
 }

crates/collab/src/tests/test_server.rs

@@ -679,8 +679,6 @@ impl TestServer {
                 stripe_api_key: None,
                 stripe_price_id: None,
                 supermaven_admin_api_key: None,
-                runpod_api_key: None,
-                runpod_api_summary_url: None,
                 user_backfiller_github_access_token: None,
             },
         })