Add configs from 143530

huggingface · Jul 15, 2024 · 132cd8f · 132cd8f
1 parent 9ef6766
commit 132cd8f
Showing 1 changed file with 20 additions and 5 deletions.
diff --git a/optimum/intel/openvino/configuration.py b/optimum/intel/openvino/configuration.py
@@ -33,10 +33,9 @@
 logger = logging.getLogger(__name__)
 
 _DEFAULT_4BIT_CONFIGS = {
-    "databricks/dolly-v2-3b": {"bits": 4, "sym": False, "group_size": 128, "ratio": 0.8},
+    "databricks/dolly-v2-3b": {"bits": 4, "sym": False, "group_size": 128, "scale_estimation": True},
     "EleutherAI/gpt-j-6b": {"bits": 4, "sym": False, "group_size": 64},
     "facebook/opt-6.7b": {"bits": 4, "sym": False, "group_size": 64, "ratio": 0.8},
-    "bigscience/bloomz-7b1": {"bits": 4, "sym": False, "group_size": 32, "ratio": 0.6},
     "togethercomputer/RedPajama-INCITE-7B-Instruct": {"bits": 4, "sym": False, "group_size": 128},
     "HuggingFaceH4/zephyr-7b-beta": {
         "bits": 4,
@@ -69,7 +68,7 @@
     "pansophic/rocket-3B": {"bits": 4, "sym": True, "group_size": 128, "ratio": 0.8},
     "THUDM/chatglm2-6b": {"bits": 4, "sym": True, "group_size": 128, "ratio": 0.72},
     "Qwen/Qwen-7B-Chat": {"bits": 4, "sym": True, "group_size": 128, "ratio": 0.6},
-    "openlm-research/open_llama_3b": {"bits": 4, "sym": True, "group_size": 64, "all_layers": True},
+    "openlm-research/open_llama_3b": {"bits": 4, "sym": False, "group_size": 64, "all_layers": True},
     "openlm-research/open_llama_3b_v2": {"bits": 4, "sym": True, "group_size": 64, "all_layers": True},
     "tiiuae/falcon-7b-instruct": {"bits": 4, "sym": True, "group_size": 64, "all_layers": True},
     "psmathur/orca_mini_3b": {
@@ -90,7 +89,12 @@
     },
     "mistralai/Mixtral-8x7B-v0.1": {"bits": 4, "sym": True, "group_size": 128, "ratio": 0.8},
     "facebook/opt-2.7b": {"bits": 4, "sym": True, "group_size": 128, "ratio": 0.7},
-    "togethercomputer/RedPajama-INCITE-Chat-3B-v1": {"bits": 4, "sym": False, "group_size": 128, "ratio": 0.8},
+    "togethercomputer/RedPajama-INCITE-Chat-3B-v1": {
+        "bits": 4,
+        "sym": False,
+        "group_size": 128,
+        "scale_estimation": True,
+    },
     "lmsys/vicuna-7b-v1.5": {"bits": 4, "sym": False, "group_size": 128, "ratio": 1.0},
     "stabilityai/stablelm-tuned-alpha-3b": {"bits": 4, "sym": False, "group_size": 128, "ratio": 0.8},
     "mistralai/Mistral-7B-v0.1": {"bits": 4, "sym": True, "group_size": 128, "ratio": 0.9},
@@ -100,7 +104,18 @@
         "group_size": 128,
         "ratio": 0.8,
         "dataset": "wikitext2",
-        "awq": True,
+        "quant_method": "awq",
+    },
+    "openai-community/gpt2": {"bits": 4, "sym": False, "group_size": 128, "ratio": 0.5, "scale_estimation": True},
+    "lmsys/longchat-7b-16k": {"bits": 4, "sym": False, "group_size": 128, "ratio": 0.9},
+    "bigcode/starcoder2-3b": {"bits": 4, "sym": False, "group_size": 128, "ratio": 0.9},
+    "TinyLlama/TinyLlama-1.1B-Chat-v1.0": {"bits": 4, "sym": False, "group_size": 128, "ratio": 0.8},
+    "stabilityai/stablelm-tuned-alpha-7b": {
+        "bits": 4,
+        "sym": False,
+        "group_size": 128,
+        "ratio": 0.6,
+        "scale_estimation": True,
     },
 }