Upload folder using huggingface_hub

README.md

@@ -0,0 +1,3 @@
+# AWQ Cache
+
+Mapping HF model names to the names as saved in the TitanML directory. 

awq_cache.json

@@ -0,0 +1,87 @@
+{
+    "llama-2-7b": {
+        "repo": "TitanML/llama2-7B-base-4bit-AWQ"
+    },
+    "llama-2-7b-hf": {
+        "repo": "TitanML/llama2-7B-base-4bit-AWQ"
+    },
+    "meta-llama/llama-2-7b": {
+        "repo": "TitanML/llama2-7B-base-4bit-AWQ"
+    },
+    "meta-llama/llama-2-7b-hf": {
+        "repo": "TitanML/llama2-7B-base-4bit-AWQ"
+    },
+    "llama-2-13b": {
+        "repo": "TitanML/llama2-13B-base-4bit-AWQ"
+    },
+    "llama-2-13b-hf": {
+        "repo": "TitanML/llama2-13B-base-4bit-AWQ"
+    },
+    "meta-llama/llama-2-13b": {
+        "repo": "TitanML/llama2-13B-base-4bit-AWQ"
+    },
+    "meta-llama/llama-2-13b-hf": {
+        "repo": "TitanML/llama2-13B-base-4bit-AWQ"
+    },
+    "llama-2-7b-chat": {
+        "repo": "TitanML/llama2-7B-chat-4bit-AWQ"
+    },
+    "llama-2-7b-chat-hf": {
+        "repo": "TitanML/llama2-7B-chat-4bit-AWQ"
+    },
+    "meta-llama/llama-2-7b-chat": {
+        "repo": "TitanML/llama2-7B-chat-4bit-AWQ"
+    },
+    "meta-llama/llama-2-7b-chat-hf": {
+        "repo": "TitanML/llama2-7B-chat-4bit-AWQ"
+    },
+    "llama-2-13b-chat": {
+        "repo": "TitanML/llama2-13B-chat-4bit-AWQ"
+    },
+    "llama-2-13b-chat-hf": {
+        "repo": "TitanML/llama2-13B-chat-4bit-AWQ"
+    },
+    "meta-llama/llama-2-13b-chat": {
+        "repo": "TitanML/llama2-13B-chat-4bit-AWQ"
+    },
+    "meta-llama/llama-2-13b-chat-hf": {
+        "repo": "TitanML/llama2-13B-chat-4bit-AWQ"
+    },
+    "vicuna-13b": {
+        "repo": "TitanML/vicuna-13B-base-4bit-AWQ"
+    },
+    "vicuna-7b": {
+        "repo": "TitanML/vicuna-7B-base-4bit-AWQ"
+    },
+    "mpt-7b": {
+        "repo": "TitanML/mpt-7b-base-8k-4bit-AWQ"
+    },
+    "mosaicml/mpt-7b": {
+        "repo": "TitanML/mpt-7b-base-8k-4bit-AWQ"
+    },
+    "mpt-7b-chat": {
+        "repo": "TitanML/mpt-7b-chat-8k-4bit-AWQ"
+    },
+    "mosaicml/mpt-7b-chat": {
+        "repo": "TitanML/mpt-7b-chat-8k-4bit-AWQ"
+    },
+    "mpt-7b-instruct": {
+        "repo": "TitanML/mpt-7b-instruct-8k-4bit-AWQ"
+    },
+    "mosaicml/mpt-7b-instruct": {
+        "repo": "TitanML/mpt-7b-instruct-8k-4bit-AWQ"
+    },
+    "opt-13b": {
+        "repo": "TitanML/opt-13b-base-4bit-AWQ"
+    },
+    "facebook/opt-13b": {
+        "repo": "TitanML/opt-13b-base-4bit-AWQ"
+    },
+    "opt-30b": {
+        "repo": "TitanML/opt-30b-base-4bit-AWQ"
+    },
+    "facebook/opt-30b": {
+        "repo": "TitanML/opt-30b-base-4bit-AWQ"
+    }
+}
+

ct2_cache.json

@@ -0,0 +1,170 @@
+{
+    "tiiuae/falcon-7b": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-falcon-7b"
+    },
+    "falcon-7b": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-falcon-7b"
+    },
+    "mosaicml/mpt-7b": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-mtb-7b"
+    },
+    "mpt-7b": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-mtb-7b"
+    },
+    "mosaicml/mpt-7b-storywriter": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-mtb-7b-storywriter"
+    },
+    "mpt-7b-storywriter": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-mtb-7b-storywriter"
+    },
+    "mosaicml/mpt-7b-chat": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-mtb-7b-chat"
+    },
+    "mpt-7b-chat": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-mtb-7b-chat"
+    },
+    "mosaicml/mpt-7b-8k": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-mtb-7b-8k"
+    },
+    "mpt-7b-8k": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-mtb-7b-8k"
+    },
+    "mosaicml/mpt-7b-8k-instruct": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-mtb-7b-8k-instruct"
+    },
+    "mpt-7b-8k-instruct": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-mtb-7b-8k-instruct"
+    },
+    "mosaicml/mpt-7b-8k-chat": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-mtb-7b-8k-chat"
+    },
+    "mpt-7b-8k-chat": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-mtb-7b-8k-chat"
+    },
+    "togethercomputer/redpajama-incite-7b-base": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-redpajama-7b-base"
+    },
+    "redpajama-incite-7b-base": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-redpajama-7b-base"
+    },
+    "togethercomputer/redpajama-incite-7b-instruct": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-redpajama-7b-instruct"
+    },
+    "redpajama-incite-7b-instruct": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-redpajama-7b-instruct"
+    },
+    "togethercomputer/redpajama-incite-7b-chat": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-redpajama-7b-chat"
+    },
+    "redpajama-incite-7b-chat": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-redpajama-7b-chat"
+    },
+    "meta-llama/llama-2-7b-chat-hf": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-llama-2-7b-chat"
+    },
+    "meta-llama/llama-2-7b-chat": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-llama-2-7b-chat"
+    },
+    "llama-2-7b-chat-hf": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-llama-2-7b-chat"
+    },
+    "llama-2-7b-chat": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-llama-2-7b-chat"
+    },
+    "google/flan-t5-xl": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-flan-xl"
+    },
+    "flan-t5-xl": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-flan-xl"
+    },
+    "conceptofmind/flan-open-llama-7b": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-flan-open-llama-7b"
+    },
+    "flan-open-llama-7b": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-flan-open-llama-7b"
+    },
+    "openlm-research/open_llama_7b": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-open-llama-7b"
+    },
+    "open_llama_7b": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-open-llama-7b"
+    },
+    "openlm-research/open_llama_7b_v2": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-open-llama-7b-v2"
+    },
+    "open_llama_7b_v2": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-open-llama-7b-v2"
+    },
+    "salesforce/xgen-7b-8k-base": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-xgen-7b-8k-base"
+    },
+    "xgen-7b-8k-base": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-xgen-7b-8k-base"
+    },
+    "salesforce/xgen-7b-8k-inst": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-xgen-7b-8k-inst"
+    },
+    "xgen-7b-8k-inst": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-xgen-7b-8k-inst"
+    },
+    "salesforce/codegen25-7b-instruct": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-codegen25-7b-instruct"
+    },
+    "codegen25-7b-instruct": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-codegen25-7b-instruct"
+    },
+    "stabilityai/stablelm-tuned-alpha-7b": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-stablelm-7b"
+    },
+    "stablelm-tuned-alpha-7b": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-stablelm-7b"
+    },
+    "tiiuae/falcon-7b-instruct": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-falcon-7b-instruct"
+    },
+    "falcon-7b-instruct": {
+        "quant_type": "int8",
+        "repo": "TitanML/ct2-int8-falcon-7b-instruct"
+    }
+}

ggml_cache.json

@@ -0,0 +1,176 @@
+{
+  "llama-2-7b": {
+    "repo": "TheBloke/Llama-2-7B-GGML",
+    "filename": "llama-2-7b.ggmlv3.q4_0.bin",
+    "model_type": "llama",
+    "tokenizer": "llama-local"
+  },
+  "meta-llama/llama-2-7b": {
+    "repo": "TheBloke/Llama-2-7B-GGML",
+    "filename": "llama-2-7b.ggmlv3.q4_0.bin",
+    "model_type": "llama",
+    "tokenizer": "llama-local"
+  },
+  "llama-2-7b-hf": {
+    "repo": "TheBloke/Llama-2-7B-GGML",
+    "filename": "llama-2-7b.ggmlv3.q4_0.bin",
+    "model_type": "llama",
+    "tokenizer": "llama-local"
+  },
+  "meta-llama/llama-2-7b-hf": {
+    "repo": "TheBloke/Llama-2-7B-GGML",
+    "filename": "llama-2-7b.ggmlv3.q4_0.bin",
+    "model_type": "llama",
+    "tokenizer": "llama-local"
+  },
+  "llama-2-7b-chat": {
+    "repo": "TheBloke/Llama-2-7B-Chat-GGML",
+    "filename": "llama-2-7b-chat.ggmlv3.q4_0.bin",
+    "model_type": "llama",
+    "tokenizer": "llama-local"
+  },
+  "meta-llama/llama-2-7b-chat": {
+    "repo": "TheBloke/Llama-2-7B-Chat-GGML",
+    "filename": "llama-2-7b-chat.ggmlv3.q4_0.bin",
+    "model_type": "llama",
+    "tokenizer": "llama-local"
+  },
+  "llama-2-7b-chat-hf": {
+    "repo": "TheBloke/Llama-2-7B-Chat-GGML",
+    "filename": "llama-2-7b-chat.ggmlv3.q4_0.bin",
+    "model_type": "llama",
+    "tokenizer": "llama-local"
+  },
+  "meta-llama/llama-2-7b-chat-hf": {
+    "repo": "TheBloke/Llama-2-7B-Chat-GGML",
+    "filename": "llama-2-7b-chat.ggmlv3.q4_0.bin",
+    "model_type": "llama",
+    "tokenizer": "llama-local"
+  },
+  "llama-2-13b": {
+    "repo": "TheBloke/Llama-2-13B-GGML",
+    "filename": "llama-2-13b.ggmlv3.q4_0.bin",
+    "model_type": "llama",
+    "tokenizer": "llama-local"
+  },
+  "meta-llama/llama-2-13b": {
+    "repo": "TheBloke/Llama-2-13B-GGML",
+    "filename": "llama-2-13b.ggmlv3.q4_0.bin",
+    "model_type": "llama",
+    "tokenizer": "llama-local"
+  },
+  "llama-2-13b-hf": {
+    "repo": "TheBloke/Llama-2-13B-GGML",
+    "filename": "llama-2-13b.ggmlv3.q4_0.bin",
+    "model_type": "llama",
+    "tokenizer": "llama-local"
+  },
+  "meta-llama/llama-2-13b-hf": {
+    "repo": "TheBloke/Llama-2-13B-GGML",
+    "filename": "llama-2-13b.ggmlv3.q4_0.bin",
+    "model_type": "llama",
+    "tokenizer": "llama-local"
+  },
+  "llama-2-13b-chat": {
+    "repo": "TheBloke/Llama-2-13B-Chat-GGML",
+    "filename": "llama-2-13b-chat.ggmlv3.q4_0.bin",
+    "model_type": "llama",
+    "tokenizer": "llama-local"
+  },
+  "meta-llama/llama-2-13b-chat": {
+    "repo": "TheBloke/Llama-2-13B-Chat-GGML",
+    "filename": "llama-2-13b-chat.ggmlv3.q4_0.bin",
+    "model_type": "llama",
+    "tokenizer": "llama-local"
+  },
+  "llama-2-13b-chat-hf": {
+    "repo": "TheBloke/Llama-2-13B-Chat-GGML",
+    "filename": "llama-2-13b-chat.ggmlv3.q4_0.bin",
+    "model_type": "llama",
+    "tokenizer": "llama-local"
+  },
+  "meta-llama/llama-2-13b-chat-hf": {
+    "repo": "TheBloke/Llama-2-13B-Chat-GGML",
+    "filename": "llama-2-13b-chat.ggmlv3.q4_0.bin",
+    "model_type": "llama",
+    "tokenizer": "llama-local"
+  },
+  "llama-2-70b": {
+    "repo": "TheBloke/Llama-2-70B-GGML",
+    "filename": "llama-2-70b.ggmlv3.q4_0.bin",
+    "model_type": "llama",
+    "tokenizer": "llama-local"
+  },
+  "meta-llama/llama-2-70b": {
+    "repo": "TheBloke/Llama-2-70B-GGML",
+    "filename": "llama-2-70b.ggmlv3.q4_0.bin",
+    "model_type": "llama",
+    "tokenizer": "llama-local"
+  },
+  "llama-2-70b-hf": {
+    "repo": "TheBloke/Llama-2-70B-GGML",
+    "filename": "llama-2-70b.ggmlv3.q4_0.bin",
+    "model_type": "llama",
+    "tokenizer": "llama-local"
+  },
+  "meta-llama/llama-2-70b-hf": {
+    "repo": "TheBloke/Llama-2-70B-GGML",
+    "filename": "llama-2-70b.ggmlv3.q4_0.bin",
+    "model_type": "llama",
+    "tokenizer": "llama-local"
+  },
+  "llama-2-70b-chat": {
+    "repo": "TheBloke/Llama-2-70B-Chat-GGML",
+    "filename": "llama-2-70b-chat.ggmlv3.q4_0.bin",
+    "model_type": "llama",
+    "tokenizer": "llama-local"
+  },
+  "meta-llama/llama-2-70b-chat": {
+    "repo": "TheBloke/Llama-2-70B-Chat-GGML",
+    "filename": "llama-2-70b-chat.ggmlv3.q4_0.bin",
+    "model_type": "llama",
+    "tokenizer": "llama-local"
+  },
+  "llama-2-70b-chat-hf": {
+    "repo": "TheBloke/Llama-2-70B-Chat-GGML",
+    "filename": "llama-2-70b-chat.ggmlv3.q4_0.bin",
+    "model_type": "llama",
+    "tokenizer": "llama-local"
+  },
+  "meta-llama/llama-2-70b-chat-hf": {
+    "repo": "TheBloke/Llama-2-70B-Chat-GGML",
+    "filename": "llama-2-70b-chat.ggmlv3.q4_0.bin",
+    "model_type": "llama",
+    "tokenizer": "llama-local"
+  },
+  "falcon-7b": {
+    "repo": "TheBloke/falcon-7b-instruct-GGML",
+    "filename": "falcon-7b-instruct.ggccv1.q4_0.bin",
+    "model_type": "falcon",
+    "tokenizer": "tiiuae/falcon-7b"
+  },
+  "tiiuae/falcon-7b": {
+    "repo": "TheBloke/falcon-7b-instruct-GGML",
+    "filename": "falcon-7b-instruct.ggccv1.q4_0.bin",
+    "model_type": "falcon",
+    "tokenizer": "tiiuae/falcon-7b"
+  },
+  "falcon-40b": {
+    "repo": "TheBloke/falcon-40b-instruct-GGML",
+    "filename": "falcon-40b-instruct.ggccv1.q4_0.bin",
+    "model_type": "falcon",
+    "tokenizer": "tiiuae/falcon-7b"
+  },
+  "ttiuae/falcon-40b": {
+    "repo": "TheBloke/falcon-40b-instruct-GGML",
+    "filename": "falcon-40b-instruct.ggccv1.q4_0.bin",
+    "model_type": "falcon",
+    "tokenizer": "tiiuae/falcon-7b"
+  },
+  "gpt2": {
+    "repo": "marella/gpt-2-ggml",
+    "filename": "ggml-model.bin",
+    "model_type": "gpt2",
+    "tokenizer": "gpt2"
+  }
+}

name_funnel.json

@@ -0,0 +1,45 @@
+{
+    "llama-2-7b": "llama-2-7b",
+    "llama-2-7b-hf": "llama-2-7b",
+    "meta-llama/llama-2-7b": "llama-2-7b",
+    "meta-llama/llama-2-7b-hf": "llama-2-7b",
+    "llama-2-7b-chat": "llama-2-7b-chat",
+    "llama-2-7b-chat-hf": "llama-2-7b-chat",
+    "meta-llama/llama-2-7b-chat": "llama-2-7b-chat",
+    "meta-llama/llama-2-7b-chat-hf": "llama-2-7b-chat",
+    "llama-2-13b": "llama-2-13b",
+    "llama-2-13b-hf": "llama-2-13b",
+    "meta-llama/llama-2-13b": "llama-2-13b",
+    "meta-llama/llama-2-13b-hf": "llama-2-13b",
+    "llama-2-13b-chat": "llama-2-13b-chat",
+    "llama-2-13b-chat-hf": "llama-2-13b-chat",
+    "meta-llama/llama-2-13b-chat": "llama-2-13b-chat",
+    "meta-llama/llama-2-13b-chat-hf": "llama-2-13b-chat",
+    "llama-2-70b": "llama-2-70b",
+    "llama-2-70b-hf": "llama-2-70b",
+    "meta-llama/llama-2-70b": "llama-2-70b",
+    "meta-llama/llama-2-70b-hf": "llama-2-70b",
+    "llama-2-70b-chat": "llama-2-70b-chat",
+    "llama-2-70b-chat-hf": "llama-2-70b-chat",
+    "meta-llama/llama-2-70b-chat": "llama-2-70b-chat",
+    "meta-llama/llama-2-70b-chat-hf": "llama-2-70b-chat",
+    "falcon-7b": "falcon-7b",
+    "tiiuae/falcon-7b": "falcon-7b",
+    "falcon-40b": "falcon-40b",
+    "tiiuae/falcon-40b": "falcon-40b",
+    "gpt2": "gpt2",
+    "vicuna-7b": "vicuna-7b",
+    "lmsys/vicuna-7b-v1.5": "vicuna-7b",
+    "vicuna-13b": "vicuna-13b",
+    "lmsys/vicuna-13b-v1.5": "vicuna-13b",
+    "facebook/opt-13b": "opt-13b",
+    "opt-13b": "opt-13b",
+    "facebook/opt-30b": "opt-30b",
+    "opt-30b": "opt-30b",
+    "mpt-7b": "mpt-7b",
+    "mosaicml/mpt-7b": "mpt-7b",
+    "mpt-7b-chat": "mpt-7b-chat",
+    "mosaicml/mpt-7b-chat": "mpt-7b-chat",
+    "mpt-7b-instruct": "mpt-7b-instruct",
+    "mosaicml/mpt-7b-instruct": "mpt-7b-instruct",
+}