From 81be5f6db82d86265611a42ae69f1a4b355d63a7 Mon Sep 17 00:00:00 2001
From: HackWeasel <david@gtedge.ai>
Date: Tue, 16 Dec 2025 09:06:59 -0500
Subject: [PATCH] fix: correct max_tokens for Groq models
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

🤖 Generated with [Claude Code](https://claude.com/claude-code)
---
 scripts/lib/migrations.sh                     | 12 +++++++
 .../010_update_model_context_windows.sql      |  2 +-
 .../migrations/028_fix_groq_max_tokens.sql    | 33 +++++++++++++++++++
 .../unified/05-create-test-data.sql           |  4 +--
 4 files changed, 48 insertions(+), 3 deletions(-)
 create mode 100644 scripts/migrations/028_fix_groq_max_tokens.sql

diff --git a/scripts/lib/migrations.sh b/scripts/lib/migrations.sh
index 0502a19..a38ac3c 100755
--- a/scripts/lib/migrations.sh
+++ b/scripts/lib/migrations.sh
@@ -305,6 +305,15 @@ check_migration_027() {
     [ "$missing" != "0" ] && [ -n "$missing" ]
 }
 
+check_migration_028() {
+    # Returns true (needs migration) if any model has wrong max_tokens
+    local wrong_value=$(docker exec gentwo-controlpanel-postgres psql -U postgres -d gt2_admin -tAc \
+        "SELECT COUNT(*) FROM model_configs WHERE
+         (model_id = 'llama-3.1-8b-instant' AND max_tokens != 32000) OR
+         (model_id = 'meta-llama/llama-guard-4-12b' AND max_tokens != 1024);" 2>/dev/null || echo "0")
+    [ "$wrong_value" != "0" ] && [ -n "$wrong_value" ]
+}
+
 # Tenant migration checks
 check_migration_T001() {
     local exists=$(docker exec gentwo-tenant-postgres-primary psql -U postgres -d gt2_tenants -tAc \
@@ -411,6 +420,9 @@ run_admin_migrations() {
     # This fixes partial 021 migrations where models were added but not assigned
     run_admin_migration "027" "scripts/migrations/027_assign_nvidia_models_to_tenants.sql" "check_migration_027" || return 1
 
+    # Migration 028: Fix Groq model max_tokens (llama-3.1-8b-instant and llama-guard)
+    run_admin_migration "028" "scripts/migrations/028_fix_groq_max_tokens.sql" "check_migration_028" || return 1
+
     log_success "All admin migrations complete"
     return 0
 }
diff --git a/scripts/migrations/010_update_model_context_windows.sql b/scripts/migrations/010_update_model_context_windows.sql
index fce9bc7..2a10ac4 100644
--- a/scripts/migrations/010_update_model_context_windows.sql
+++ b/scripts/migrations/010_update_model_context_windows.sql
@@ -11,7 +11,7 @@ BEGIN
     -- LLaMA 3.1 8B Instant
     UPDATE model_configs
     SET context_window = 131072,
-        max_tokens = 131072,
+        max_tokens = 32000,
         updated_at = NOW()
     WHERE model_id = 'llama-3.1-8b-instant'
     AND (context_window IS NULL OR max_tokens IS NULL);
diff --git a/scripts/migrations/028_fix_groq_max_tokens.sql b/scripts/migrations/028_fix_groq_max_tokens.sql
new file mode 100644
index 0000000..498d3ac
--- /dev/null
+++ b/scripts/migrations/028_fix_groq_max_tokens.sql
@@ -0,0 +1,33 @@
+-- Migration 028: Fix Groq Model Max Tokens
+-- Corrects max_tokens for models that had incorrect values in test data
+
+DO $$
+DECLARE
+    updated_count INTEGER := 0;
+BEGIN
+    -- LLaMA 3.1 8B Instant: max_tokens should be 32000 (was incorrectly 131072)
+    UPDATE model_configs
+    SET max_tokens = 32000,
+        updated_at = NOW()
+    WHERE model_id = 'llama-3.1-8b-instant'
+    AND max_tokens != 32000;
+
+    GET DIAGNOSTICS updated_count = ROW_COUNT;
+    IF updated_count > 0 THEN
+        RAISE NOTICE 'Updated % records for llama-3.1-8b-instant max_tokens -> 32000', updated_count;
+    END IF;
+
+    -- LLaMA Guard 4 12B: max_tokens should be 1024 (was incorrectly 8192 in test data)
+    UPDATE model_configs
+    SET max_tokens = 1024,
+        updated_at = NOW()
+    WHERE model_id = 'meta-llama/llama-guard-4-12b'
+    AND max_tokens != 1024;
+
+    GET DIAGNOSTICS updated_count = ROW_COUNT;
+    IF updated_count > 0 THEN
+        RAISE NOTICE 'Updated % records for llama-guard-4-12b max_tokens -> 1024', updated_count;
+    END IF;
+
+    RAISE NOTICE 'Migration 028 complete: Groq max_tokens corrected';
+END $$;
diff --git a/scripts/postgresql/unified/05-create-test-data.sql b/scripts/postgresql/unified/05-create-test-data.sql
index 1dc1ff3..2ad43e2 100644
--- a/scripts/postgresql/unified/05-create-test-data.sql
+++ b/scripts/postgresql/unified/05-create-test-data.sql
@@ -94,7 +94,7 @@ INSERT INTO public.model_configs (
 -- Groq Llama 3.1 8B Instant (fast, cheap)
 ('llama-3.1-8b-instant', 'Groq Llama 3.1 8b Instant', '1.0', 'groq', 'llm',
  'https://api.groq.com/openai/v1/chat/completions',
- 131072, 131072,
+ 131072, 32000,
  '{"reasoning": false, "function_calling": false, "vision": false, "audio": false, "streaming": false, "multilingual": false}'::json,
  0.05, 0.08, true, 'unknown', 0, 0, 100, 0,
  '{"global_access": true}'::json, '[]'::json,
@@ -148,7 +148,7 @@ INSERT INTO public.model_configs (
 -- Groq Llama Guard 4 12B (safety/moderation model)
 ('meta-llama/llama-guard-4-12b', 'Groq Llama Guard 4 12B', '1.0', 'groq', 'llm',
  'https://api.groq.com/openai/v1/chat/completions',
- 131072, 8192,
+ 131072, 1024,
  '{"reasoning": false, "function_calling": false, "vision": false, "audio": false, "streaming": false, "multilingual": false}'::json,
  0.20, 0.20, true, 'unknown', 0, 0, 100, 0,
  '{"global_access": true}'::json, '[]'::json,