initial release

.gitattributes

@@ -0,0 +1,3 @@
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.json filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,86 @@
+---
+language:
+  - en
+thumbnail: null
+tags:
+  - text generation
+  - conversational
+pipeline_tag: text-generation
+inference: false
+---
+<h1 style="text-align: center">Pygmalion 13B</h1>
+<h2 style="text-align: center">A conversational LLaMA fine-tune.</h2>
+
+## Model Details
+
+Pygmalion 13B is a dialogue model based on Meta's LLaMA-13B.
+
+This is version 1. It has been fine-tuned using a subset of the data from Pygmalion-6B-v8-pt4, for those of you familiar with the project.
+
+## Applying the XORs
+
+**The model weights in this repository cannot be used as-is.** The files here are XORs due to licensing concerns. To obtain proper, usable model weights you need to:
+
+- Request access to the original LLaMA weights from Meta [through this form](https://docs.google.com/forms/d/e/1FAIpQLSfqNECQnMkycAp2jP4Z9TFX0cGR4uf7b_fBxjY_OjhJILlKGA/viewform?usp=send_form)
+- Convert them to the HuggingFace Transformers format by using the [convert_llama_weights_to_hf.py](https://github.com/huggingface/transformers/blob/849367ccf741d8c58aa88ccfe1d52d8636eaf2b7/src/transformers/models/llama/convert_llama_weights_to_hf.py) script **for your version of the `transformers` library**
+- With the LLaMA-13B weights in hand, you can use the [xor_codec.py](./xor_codec.py) script provided in this repository:
+
+  ```bash
+  python3 xor_codec.py \
+    ./pygmalion-13b \
+    ./xor_encoded_files \
+    /path/to/hf-converted/llama-13b \
+    --decode
+  ```
+
+For reference, these are the hashes you should get after following the steps above:
+
+```bash
+$ rhash --sha256 *
+3b12e6740652990ac386b6136119aaca698aa547d9460e1ef243a7d17d489fe3  config.json
+e14c4af01ea4febe3448d9db29c6dbd982966c5161a31c5185b8fe6d6796509a  generation_config.json
+6b05c8e8ae9c1065e4c7cfd2b61311191d2ad5735a2e4beab98fc53b49375af8  pytorch_model-00001-of-00003.bin
+4cd096ac310b6bbc3acd3d729277427ad3c3d5740619462dc8f907dfeac3e66f  pytorch_model-00002-of-00003.bin
+a2fe9ac5d7005e65c58b8d14818678dd0730689f518612b90cca19fed7c483ad  pytorch_model-00003-of-00003.bin
+72e91e29282dae48ea5562fcf4d6ca0d5a9c2a30ebc8d67174a19e192552a20b  pytorch_model.bin.index.json
+ff3b4a612c4e447acb02d40071bddd989fe0da87eb5b7fe0dbadfc4f74de7531  special_tokens_map.json
+f9ffc4aede0845ab65324ce5dccb823dca2427f9a0710981e5bc2398d73d8162  tokenizer.json
+9e556afd44213b6bd1be2b850ebbbd98f5481437a8021afaf58ee7fb1818d347  tokenizer.model
+380608719f3af6ef2b343e2ed53bf55556678609337e88a14f58cc49177b9e18  tokenizer_config.json
+```
+
+## Prompting
+
+The model was trained on the usual Pygmalion persona + chat format, so any of the usual UIs should already handle everything correctly. If you're using the model directly, this is the expected formatting:
+
+```
+[CHARACTER]'s Persona: [A few sentences about the character you want the model to play]
+<START>
+[DIALOGUE HISTORY]
+You: [User's input message here]
+[CHARACTER]:
+```
+
+Where `[CHARACTER]` is, as you can probably guess, the name of the character you want the model to portray, `<START>` should be used verbatim as a delimiter token to separate persona and scenario data from the dialogue, and `[DIALOGUE HISTORY]` is a sliding window of chat history so the model can have conversational context to draw from. Here's a concrete example:
+
+```
+Assistant's Persona: Assistant is a highly intelligent language model trained to comply with user requests.
+<START>
+Assistant: Hello! How may I help you today?
+You: What is Zork?
+Assistant:
+```
+
+Which will generate something like:
+
+```
+ Zork is an interactive fiction computer game created in the 1970s by Infocom, Inc., which was later acquired by Activision Blizzard. It is widely considered one of the most influential games ever made and has been credited with popularizing text-based adventure games. The original version of Zork was written in the programming language MACRO-10, but it was ported to many other platforms over the years."
+```
+
+The model will automatically emit an end-of-text token (`</s>`) when it judges that the response is complete.
+
+## Limitations and biases
+
+The intended use-case for this model is fictional conversation for entertainment purposes. Any other sort of usage is out of scope.
+
+As such, it was **not** fine-tuned to be safe and harmless: the base model _and_ this fine-tune have been trained on data known to contain profanity and texts that are lewd or otherwise offensive. It may produce socially unacceptable or undesirable text, even if the prompt itself does not include anything explicitly offensive. Outputs might often be factually wrong or misleading.

xor_codec.py

@@ -0,0 +1,92 @@
+'''
+Adapted from OpenAssistant's original xor_codec.py:
+https://huggingface.co/OpenAssistant/oasst-sft-6-llama-30b-xor/raw/main/xor_codec.py
+'''
+import os
+import sys
+import shutil
+import gzip
+import numpy
+from pathlib import Path
+
+def xor_uncompressed(dst, src_payload, src_base, block_size=4096):
+    fp_payload = open(src_payload, 'rb')
+    fp_base = open(src_base, 'rb')
+    with open(dst, 'wb') as fp:
+        while True:
+            buf1 = numpy.array(bytearray(fp_payload.read(block_size)), dtype=numpy.uint8)
+            buf2 = numpy.array(bytearray(fp_base.read(block_size)), dtype=numpy.uint8)
+            padding = len(buf1) - len(buf2)
+            if padding > 0: buf2 = numpy.pad(buf2, (0, padding), 'constant', constant_values=(0,))
+            if padding < 0: buf2 = buf2[:len(buf1)]
+            buf = numpy.bitwise_xor(buf1, buf2)
+            fp.write(buf)
+            if len(buf1) < block_size: break
+    fp_payload.close()
+    fp_base.close()
+
+def xor_encode(dst, src_payload, src_base, block_size=4096):
+    fp_payload = open(src_payload, 'rb')
+    fp_base = open(src_base, 'rb')
+    with gzip.open(dst, 'wb') as fp:
+        while True:
+            buf1 = numpy.array(bytearray(fp_payload.read(block_size)), dtype=numpy.uint8)
+            buf2 = numpy.array(bytearray(fp_base.read(block_size)), dtype=numpy.uint8)
+            padding = len(buf1) - len(buf2)
+            if padding > 0: buf2 = numpy.pad(buf2, (0, padding), 'constant', constant_values=(0,))
+            if padding < 0: buf2 = buf2[:len(buf1)]
+            buf = numpy.bitwise_xor(buf1, buf2)
+            fp.write(buf)
+            if len(buf1) < block_size: break
+    fp_payload.close()
+    fp_base.close()
+
+def xor_decode(dst, src_payload, src_base, block_size=4096):
+    fp_payload = gzip.open(src_payload, 'rb')
+    fp_base = open(src_base, 'rb')
+    with open(dst, 'wb') as fp:
+        while True:
+            buf1 = numpy.array(bytearray(fp_payload.read(block_size)), dtype=numpy.uint8)
+            buf2 = numpy.array(bytearray(fp_base.read(block_size)), dtype=numpy.uint8)
+            padding = len(buf1) - len(buf2)
+            if padding > 0: buf2 = numpy.pad(buf2, (0, padding), 'constant', constant_values=(0,))
+            if padding < 0: buf2 = buf2[:len(buf1)]
+            buf = numpy.bitwise_xor(buf1, buf2)
+            fp.write(buf)
+            if len(buf1) < block_size: break
+    fp_payload.close()
+    fp_base.close()
+
+def xor_dir(dst, src_payload, src_base, decode=True, compress=True):
+    if compress:
+        xor = xor_decode if decode else xor_encode
+    else:
+        xor = xor_uncompressed
+    Path(dst).mkdir(parents=True, exist_ok=True)
+    for path in os.listdir(src_payload):
+        # Don't care about uncopyrightable text files, just copy over.
+        if ".json" in path:
+            print("[*] Copying '%s'" % path)
+            shutil.copy(f"{src_payload}/{path}", f"{dst}/{path}")
+            continue
+
+        print("[*] Processing '%s'" % path)
+        try:
+            xor("%s/%s" % (dst, path), "%s/%s" % (src_payload, path), "%s/%s" % (src_base, path))
+        except Exception as e:
+            print("Exception when processing '%s'" % path)
+
+if __name__ == "__main__":
+    if len(sys.argv) < 4:
+        print("Usage: xor.py <DESTINATION> <PAYLOAD SOURCE> <LLAMA SOURCE> [--encode] [--compress]")
+        exit()
+    dst = sys.argv[1]
+    src_payload = sys.argv[2]
+    src_base = sys.argv[3]
+    decode = True
+    compress = False
+    if len(sys.argv) > 4:
+        for arg in sys.argv[4:]:
+            if arg == "--encode": decode = False
+            if arg == "--compress": compress = True
+    xor_dir(dst, src_payload, src_base, decode=decode, compress=compress)

xor_encoded_files/config.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3b12e6740652990ac386b6136119aaca698aa547d9460e1ef243a7d17d489fe3
+size 598

xor_encoded_files/generation_config.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e14c4af01ea4febe3448d9db29c6dbd982966c5161a31c5185b8fe6d6796509a
+size 137

xor_encoded_files/pytorch_model-00001-of-00003.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ed8d9d8d0f59e0d6205b1c5d6bc9b85b3358367be059eef840bc26328ca58179
+size 9948728622

xor_encoded_files/pytorch_model-00002-of-00003.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:551373b2bd6162a37eae8d03d7e626aa7c64a0fc70a62794ef047b266298c89e
+size 9904165088

xor_encoded_files/pytorch_model-00003-of-00003.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fdb2cc42fd5321fe993699f198c8b4fdf7b8f4e4ee1903f6428b4e073f0a0d82
+size 6178983817

xor_encoded_files/pytorch_model.bin.index.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:72e91e29282dae48ea5562fcf4d6ca0d5a9c2a30ebc8d67174a19e192552a20b
+size 33444

xor_encoded_files/special_tokens_map.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ff3b4a612c4e447acb02d40071bddd989fe0da87eb5b7fe0dbadfc4f74de7531
+size 411

xor_encoded_files/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f9ffc4aede0845ab65324ce5dccb823dca2427f9a0710981e5bc2398d73d8162
+size 1842665

xor_encoded_files/tokenizer.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:68147850c080987172d24ad27a9ba2c65c71b46e248e8ee0f0c4eda90e2ca558
+size 499723

xor_encoded_files/tokenizer_config.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:380608719f3af6ef2b343e2ed53bf55556678609337e88a14f58cc49177b9e18
+size 700