sumuks HF Staff commited on
Commit
ed6c708
·
1 Parent(s): 85cc54e

add relevant, AND cheaper models

Browse files
Files changed (1) hide show
  1. backend/config/models_config.py +17 -9
backend/config/models_config.py CHANGED
@@ -6,18 +6,25 @@ This file centralizes all configurations related to models and providers used in
6
 
7
  # Definition of preferred providers, used in get_available_model_provider.py
8
  # PREFERRED_PROVIDERS = ["sambanova", "novita"]
9
- PREFERRED_PROVIDERS = ["fireworks-ai", "sambanova", "novita"]
 
10
 
11
  # Default models to evaluate for evaluation
12
  DEFAULT_EVALUATION_MODELS = [
13
- "Qwen/QwQ-32B",
14
- "Qwen/Qwen2.5-72B-Instruct",
15
- "Qwen/Qwen2.5-32B-Instruct",
16
- "meta-llama/Llama-3.1-8B-Instruct",
17
- "meta-llama/Llama-3.3-70B-Instruct",
18
- "deepseek-ai/DeepSeek-R1-Distill-Llama-70B",
19
- "mistralai/Mistral-Small-24B-Instruct-2501",
20
  ]
 
 
 
 
 
 
 
 
 
21
 
22
  # Modèles alternatifs à utiliser si le modèle par défaut n'est pas disponible
23
  ALTERNATIVE_BENCHMARK_MODELS = [
@@ -33,7 +40,8 @@ ALTERNATIVE_BENCHMARK_MODELS = [
33
  ]
34
 
35
  # Required model for create_bench_config_file.py (only one default model)
36
- DEFAULT_BENCHMARK_MODEL = "Qwen/Qwen2.5-32B-Instruct"
 
37
 
38
  # Models by roles for benchmark configuration
39
  # All roles use the default model except chunking
 
6
 
7
  # Definition of preferred providers, used in get_available_model_provider.py
8
  # PREFERRED_PROVIDERS = ["sambanova", "novita"]
9
+ # increased number of providers
10
+ PREFERRED_PROVIDERS = ["fireworks-ai", "cerebras", "sambanova", "together", "nebius", "novita", "hyperbolic", "cohere", "hf-inference"]
11
 
12
  # Default models to evaluate for evaluation
13
  DEFAULT_EVALUATION_MODELS = [
14
+ "Qwen/Qwen3-30B-A3B",
15
+ "Qwen/Qwen3-235B-A22B",
16
+ "meta-llama/Llama-4-Maverick-17B-128E-Instruct",
17
+ "meta-llama/Llama-4-Scout-17B-16E-Instruct"
 
 
 
18
  ]
19
+ # DEFAULT_EVALUATION_MODELS = [
20
+ # "Qwen/QwQ-32B",
21
+ # "Qwen/Qwen2.5-72B-Instruct",
22
+ # "Qwen/Qwen2.5-32B-Instruct",
23
+ # "meta-llama/Llama-3.1-8B-Instruct",
24
+ # "meta-llama/Llama-3.3-70B-Instruct",
25
+ # "deepseek-ai/DeepSeek-R1-Distill-Llama-70B",
26
+ # "mistralai/Mistral-Small-24B-Instruct-2501",
27
+ # ]
28
 
29
  # Modèles alternatifs à utiliser si le modèle par défaut n'est pas disponible
30
  ALTERNATIVE_BENCHMARK_MODELS = [
 
40
  ]
41
 
42
  # Required model for create_bench_config_file.py (only one default model)
43
+ DEFAULT_BENCHMARK_MODEL = "Qwen/Qwen3-30B-A3B"
44
+ # DEFAULT_BENCHMARK_MODEL = "Qwen/Qwen2.5-32B-Instruct"
45
 
46
  # Models by roles for benchmark configuration
47
  # All roles use the default model except chunking