feat(models): add Cerebras Llama 3.1 hyphenated aliases (cerebras/llama-3.1-8b, cerebras/llama-3.1-70b) (#48157) (#48158)

litellm/model_prices_and_context_window_backup.json

@@ -15663,6 +15663,28 @@
             "/v1/images/generations"
         ]
     },
+    "cerebras/llama-3.1-70b": {
+        "input_cost_per_token": 6e-07,
+        "litellm_provider": "cerebras",
+        "max_input_tokens": 128000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 6e-07,
+        "supports_function_calling": true,
+        "supports_tool_choice": true
+    },
+    "cerebras/llama-3.1-8b": {
+        "input_cost_per_token": 1e-07,
+        "litellm_provider": "cerebras",
+        "max_input_tokens": 128000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 1e-07,
+        "supports_function_calling": true,
+        "supports_tool_choice": true
+    },
     "cerebras/llama-3.3-70b": {
         "input_cost_per_token": 8.5e-07,
         "litellm_provider": "cerebras",

model_prices_and_context_window.json

@@ -15663,6 +15663,28 @@
             "/v1/images/generations"
         ]
     },
+    "cerebras/llama-3.1-70b": {
+        "input_cost_per_token": 6e-07,
+        "litellm_provider": "cerebras",
+        "max_input_tokens": 128000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 6e-07,
+        "supports_function_calling": true,
+        "supports_tool_choice": true
+    },
+    "cerebras/llama-3.1-8b": {
+        "input_cost_per_token": 1e-07,
+        "litellm_provider": "cerebras",
+        "max_input_tokens": 128000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 1e-07,
+        "supports_function_calling": true,
+        "supports_tool_choice": true
+    },
     "cerebras/llama-3.3-70b": {
         "input_cost_per_token": 8.5e-07,
         "litellm_provider": "cerebras",

tests/unit/test_cerebras_llama_model_metadata.py

@@ -0,0 +1,115 @@
+# What does this test?
+"""
+Verifies that Cerebras Llama 3.1 models (standard hyphenated aliases):
+- cerebras/llama-3.1-8b
+- cerebras/llama-3.1-70b
+
+have correct rate card metadata, context window limits, and cost calculation parity
+with the un-hyphenated variants in model_prices_and_context_window.json.
+Fixes: https://github.com/BerriAI/litellm/issues/48157
+"""
+
+import json
+from pathlib import Path
+
+import pytest
+
+import litellm
+
+
+@pytest.fixture(scope="module")
+def model_cost_maps():
+    root = Path(__file__).resolve().parents[2]
+    main_path = root / "model_prices_and_context_window.json"
+    backup_path = root / "litellm" / "model_prices_and_context_window_backup.json"
+
+    with open(main_path, "r", encoding="utf-8") as f:
+        main_map = json.load(f)
+    with open(backup_path, "r", encoding="utf-8") as f:
+        backup_map = json.load(f)
+
+    return main_map, backup_map
+
+
+CEREBRAS_MODELS_EXPECTED = {
+    "cerebras/llama-3.1-8b": {
+        "input_cost_per_token": 1e-07,
+        "output_cost_per_token": 1e-07,
+        "max_input_tokens": 128000,
+        "max_output_tokens": 128000,
+        "mode": "chat",
+        "litellm_provider": "cerebras",
+    },
+    "cerebras/llama-3.1-70b": {
+        "input_cost_per_token": 6e-07,
+        "output_cost_per_token": 6e-07,
+        "max_input_tokens": 128000,
+        "max_output_tokens": 128000,
+        "mode": "chat",
+        "litellm_provider": "cerebras",
+    },
+}
+
+PAIR_PARITY = [
+    ("cerebras/llama-3.1-8b", "cerebras/llama3.1-8b"),
+    ("cerebras/llama-3.1-70b", "cerebras/llama3.1-70b"),
+]
+
+
+@pytest.mark.parametrize("model_id,expected_data", CEREBRAS_MODELS_EXPECTED.items())
+def test_cerebras_llama_model_metadata(model_cost_maps, model_id, expected_data):
+    """Test that hyphenated Cerebras Llama 3.1 models exist and have expected metadata in both catalogs."""
+    main_map, backup_map = model_cost_maps
+
+    assert model_id in main_map, f"Missing {model_id} in main catalog"
+    assert model_id in backup_map, f"Missing {model_id} in backup catalog"
+
+    main_entry = main_map[model_id]
+    backup_entry = backup_map[model_id]
+    assert main_entry == backup_entry, f"Parity mismatch between main and backup for {model_id}"
+
+    for key, expected_val in expected_data.items():
+        assert main_entry.get(key) == expected_val, (
+            f"{model_id}[{key}] got {main_entry.get(key)}, expected {expected_val}"
+        )
+
+
+@pytest.mark.parametrize("hyphenated,unhyphenated", PAIR_PARITY)
+def test_cerebras_llama_parity_with_unhyphenated(model_cost_maps, hyphenated, unhyphenated):
+    """Test that hyphenated and un-hyphenated variants have identical pricing and context limits."""
+    main_map, _ = model_cost_maps
+
+    assert hyphenated in main_map
+    assert unhyphenated in main_map
+
+    entry_hyphen = main_map[hyphenated]
+    entry_unhyphen = main_map[unhyphenated]
+
+    assert entry_hyphen["input_cost_per_token"] == entry_unhyphen["input_cost_per_token"]
+    assert entry_hyphen["output_cost_per_token"] == entry_unhyphen["output_cost_per_token"]
+    assert entry_hyphen["max_input_tokens"] == entry_unhyphen["max_input_tokens"]
+    assert entry_hyphen["max_output_tokens"] == entry_unhyphen["max_output_tokens"]
+
+
+def test_cerebras_llama_cost_per_token():
+    """Test that litellm.cost_per_token accurately computes costs for hyphenated Cerebras models."""
+    prompt_tokens = 1_000_000
+    completion_tokens = 1_000_000
+
+    # llama-3.1-8b: $0.10 input + $0.10 output per 1M tokens
+    cost_prompt, cost_comp = litellm.cost_per_token(
+        model="cerebras/llama-3.1-8b",
+        prompt_tokens=prompt_tokens,
+        completion_tokens=completion_tokens,
+    )
+    assert pytest.approx(cost_prompt, rel=1e-5) == 0.10
+    assert pytest.approx(cost_comp, rel=1e-5) == 0.10
+
+    # llama-3.1-70b: $0.60 input + $0.60 output per 1M tokens
+    cost_prompt_70b, cost_comp_70b = litellm.cost_per_token(
+        model="cerebras/llama-3.1-70b",
+        prompt_tokens=prompt_tokens,
+        completion_tokens=completion_tokens,
+    )
+    assert pytest.approx(cost_prompt_70b, rel=1e-5) == 0.60
+    assert pytest.approx(cost_comp_70b, rel=1e-5) == 0.60