Add files

.gitignore

@@ -0,0 +1,7 @@
+convert.py
+llama.cpp*/
+pytorch_model.bin
+*.sha
+*.tar.gz
+tokenizer.model
+config.json

Makefile

@@ -0,0 +1,42 @@
+MODEL_NAME=	open-llama-3b-v2
+PYTHON?=	python
+LLAMA_TAG=	7487137
+LLAMA_TAR=	master-$(LLAMA_TAG).tar.gz
+LLAMA_DIR=	llama.cpp-master-$(LLAMA_TAG)
+LLAMA_FLAGS=	LLAMA_NO_K_QUANTS=1
+HF_REPO=	openlm-research/open_llama_3b_v2
+HF_REF=		main
+HF_FILES=	pytorch_model.bin \
+		tokenizer.model \
+		config.json
+$(HF_FILES): 	SITE=	https://huggingface.co/$(HF_REPO)/resolve/$(HF_REF)
+$(LLAMA_TAR): 	SITE=	https://github.com/ggerganov/llama.cpp/archive/refs/tags
+
+FILES= $(HF_FILES) $(LLAMA_TAR)
+
+QUANTS=		f16 q4_0 q4_1 q5_0 q5_1 q8_0
+MODEL_FILES=	$(foreach q,$(QUANTS),$(MODEL_NAME)-$(q).bin)
+
+.PHONY: all
+all: $(MODEL_FILES) SHA256SUMS
+
+$(FILES):
+	curl -L -o $@ --url $(SITE)/$@
+
+$(LLAMA_DIR): | $(LLAMA_TAR)
+	tar -xf $(LLAMA_TAR)
+
+$(LLAMA_DIR)/quantize: | $(LLAMA_DIR)
+	$(MAKE) -C $(LLAMA_DIR) $(LLAMA_FLAGS) quantize
+
+$(MODEL_NAME)-f16.bin: $(HF_FILES) | $(LLAMA_DIR)
+	$(PYTHON) $(LLAMA_DIR)/convert.py --outtype f16 --outfile $@ .
+
+$(MODEL_NAME)-q%.bin: $(MODEL_NAME)-f16.bin $(LLAMA_DIR)/quantize
+	$(LLAMA_DIR)/quantize $< $@ q$*
+
+%.sha: %
+	sha256sum $< > $@
+
+SHA256SUMS: $(addsuffix .sha,$(MODEL_FILES))
+	cat $^ > $@

README.md

@@ -1,3 +1,25 @@
 ---
 license: apache-2.0
 ---
+
+# ggml versions of OpenLLaMa 3B v2
+
+- Version: version 2 final
+- Project: [OpenLLaMA: An Open Reproduction of LLaMA](https://github.com/openlm-research/open_llama)
+- Model: [openlm-research/open_llama_3b_v2](https://huggingface.co/openlm-research/open_llama_3b_v2)
+- [llama.cpp](https://github.com/ggerganov/llama.cpp): build 607(ffb06a3) or later
+
+## Use with llama.cpp
+
+Support is now merged to master branch.
+
+## Newer quantizations
+
+There are now more quantization types in llama.cpp, some lower than 4 bits.
+Currently these are not well supported because of technical reasons.
+If you want to use them, you have to build llama.cpp (from build 829 (ff5d58f)) with the `LLAMA_QKK_64` Make or CMake variable enabled (see PR [#2001](https://github.com/ggerganov/llama.cpp/pull/2001)).
+Then you can quantize the F16 or maybe Q8_0 version to what you want.
+
+## Perplexity on wiki.test.raw
+
+Coming soon ...

SHA256SUMS

@@ -0,0 +1,6 @@
+e89264d46a993cfc93551a90b053ffeab775d9c9ddaa5ce8d604209140fbc63d  open-llama-3b-v2-f16.bin
+06241981eab5d3a06949ad888b2da9dc4eb797e728be23e2bee7b3906454a4fe  open-llama-3b-v2-q4_0.bin
+b5176efb011e7eb78ee1d9d0c49134a3079c6ebc5e2b0bcb68fea21fbb3e2cad  open-llama-3b-v2-q4_1.bin
+da0b0d5c95579916e785441ea5b2d52cec136233c8a512d9200474f2800646aa  open-llama-3b-v2-q5_0.bin
+5f14e98356574f765374da1b8e25e7d1ced8db63ff2ed385215a94e08227ae45  open-llama-3b-v2-q5_1.bin
+789b57bc3add266ffc165f02f5f002b731e8b19aa26d194c29392de1576ce58f  open-llama-3b-v2-q8_0.bin

open-llama-3b-v2-f16.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e89264d46a993cfc93551a90b053ffeab775d9c9ddaa5ce8d604209140fbc63d
+size 6853741856

open-llama-3b-v2-q4_0.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:06241981eab5d3a06949ad888b2da9dc4eb797e728be23e2bee7b3906454a4fe
+size 1928429856

open-llama-3b-v2-q4_1.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b5176efb011e7eb78ee1d9d0c49134a3079c6ebc5e2b0bcb68fea21fbb3e2cad
+size 2142573856

open-llama-3b-v2-q5_0.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:da0b0d5c95579916e785441ea5b2d52cec136233c8a512d9200474f2800646aa
+size 2356717856

open-llama-3b-v2-q5_1.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5f14e98356574f765374da1b8e25e7d1ced8db63ff2ed385215a94e08227ae45
+size 2570861856

open-llama-3b-v2-q8_0.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:789b57bc3add266ffc165f02f5f002b731e8b19aa26d194c29392de1576ce58f
+size 3641581856