Update to latest llama.cpp

Dockerfile

@@ -21,7 +21,7 @@ RUN apt update && \
     echo "deb [ arch=amd64,arm64 ] https://repo.mongodb.org/apt/ubuntu jammy/mongodb-org/6.0 multiverse" | tee /etc/apt/sources.list.d/mongodb-org-6.0.list && \
     apt-get update && \
     apt-get install -y mongodb-org && \
-    git clone https://github.com/ggerganov/llama.cpp.git --branch master-5a5f8b1
+    git clone https://github.com/ggerganov/llama.cpp.git --branch master-ee0c40d
 
 RUN pip install --upgrade pip
 

api/pyproject.toml

@@ -64,3 +64,4 @@ beanie = "^1.17.0"
 dnspython = "^2.3.0"
 lazy-model = "^0.0.5"
 requests = "^2.28.2"
+numpy = "^1.24.2"

api/src/serge/routers/chat.py

@@ -124,7 +124,7 @@ async def event_generator():
                 prompt=full_prompt,
                 params=chat.parameters,
             ):
-                await asyncio.sleep(0.1)
+                await asyncio.sleep(0.01)
 
                 chunks.append(output)
                 full_answer += output
@@ -161,7 +161,7 @@ async def ask_a_question(chat_id: str, prompt: str):
             prompt=full_prompt,
             params=chat.parameters,
         ):
-            await asyncio.sleep(0.1)
+            await asyncio.sleep(0.01)
             answer += output
     except Exception as e:
             error = e.__str__()

api/src/serge/utils/convert.py

@@ -8,7 +8,9 @@
 import os
 import struct
 import sys
+
 from sentencepiece import SentencePieceProcessor
+from serge.utils.migrate import migrate
 
 HPARAMS = keys = ["vocab_size", "dim", "multiple_of", "n_heads", "n_layers"]
 
@@ -115,6 +117,7 @@ def convert_all(dir_model: str, tokenizer_model: str):
         tokenizer = SentencePieceProcessor(tokenizer_model)
         for file in files:
             convert_one_file(file, tokenizer)
+            migrate(file)
     except OSError:
         print("Missing tokenizer, don't forget to download it!")
 

api/src/serge/utils/generate.py

@@ -10,7 +10,7 @@ async def generate(
     prompt: str,
     params: ChatParameters
 ):
-    CHUNK_SIZE = 4
+    CHUNK_SIZE = 64
     await params.fetch_all_links()
 
     args = (

api/src/serge/utils/migrate.py

@@ -0,0 +1,305 @@
+# Migrate ggml file(s) with ggmf magic to ggml file with ggjt magic
+#
+# We caused a breaking change to the file format on 2023-03-30 in:
+#     https://github.com/ggerganov/llama.cpp/pull/613
+#
+# (1) If you still have the Meta LLaMA .pth files, then close this
+#     file now; you can just run `convert-pth-to-ggml.py` again to
+#     migrate to the new format. The tool is easier to use too. It
+#     isn't necessary anymore to manage split output files because
+#     the new format always combines things into a single file.
+#
+# (2) If you deleted the Meta LLaMA .pth files due to save on disk
+#     space, then this tool is intended to help you.  Please check
+#     out the instructions below.
+#
+# USAGE
+#
+#     python migrate-ggml-2023-03-30-pr613.py INPUT OUTPUT
+#
+# PREREQUISITES
+#
+#     pip install numpy
+#     cd llama.cpp
+#     make -j4
+#
+# EXAMPLE (7B MODEL)
+#
+#     # you can replace all the 'f16' with 'q4_0' if you're using quantized weights
+#     python migrate-ggml-2023-03-30-pr613.py models/7B/ggml-model-f16.bin models/7B/ggml-model-f16-ggjt.bin
+#
+#     # check that it works
+#     ./main -m models/7B/ggml-model-f16-ggjt.bin -p 'Question: Do you love me?'
+#
+#     # you can delete the old files
+#     rm -f models/7B/ggml-model-f16.bin
+#     mv models/7B/ggml-model-f16-ggjt.bin models/7B/ggml-model-f16.bin
+#
+# EXAMPLE (13B MODEL)
+#
+#     # you can replace all the 'f16' with 'q4_0' if you're using quantized weights
+#     python migrate-ggml-2023-03-30-pr613.py models/13B/ggml-model-f16.bin models/13B/ggml-model-f16-ggjt.bin
+#
+#     # check that it works
+#     ./main -m models/13B/ggml-model-f16-ggjt.bin -p 'Question: Do you love me?'
+#
+#     # you can delete the old files
+#     rm -f models/13B/ggml-model-f16.bin*
+#     mv models/13B/ggml-model-f16-ggjt.bin models/13B/ggml-model-f16.bin
+#
+
+import argparse
+import os
+import sys
+import json
+import struct
+import numpy as np
+
+QK = 32
+
+GGML_TYPE_Q4_0  = 0
+GGML_TYPE_Q4_1  = 1
+GGML_TYPE_I8    = 2
+GGML_TYPE_I16   = 3
+GGML_TYPE_I32   = 4
+GGML_TYPE_F16   = 5
+GGML_TYPE_F32   = 6
+
+WTYPE_NAMES = {
+    0: "F32",
+    1: "F16",
+    2: "Q4_0",
+    3: "Q4_1",
+}
+
+WTYPES = {
+    0: GGML_TYPE_F32,
+    1: GGML_TYPE_F16,
+    2: GGML_TYPE_Q4_0,
+    3: GGML_TYPE_Q4_1,
+}
+
+GGML_BLCK_SIZE = {
+    GGML_TYPE_Q4_0:  QK,
+    GGML_TYPE_Q4_1:  QK,
+    GGML_TYPE_I8:    1,
+    GGML_TYPE_I16:   1,
+    GGML_TYPE_I32:   1,
+    GGML_TYPE_F16:   1,
+    GGML_TYPE_F32:   1,
+}
+
+GGML_TYPE_SIZE = {
+    GGML_TYPE_Q4_0: 4   + QK//2,
+    GGML_TYPE_Q4_1: 4*2 + QK//2,
+    GGML_TYPE_I8:   1,
+    GGML_TYPE_I16:  2,
+    GGML_TYPE_I32:  4,
+    GGML_TYPE_F16:  2,
+    GGML_TYPE_F32:  4,
+}
+
+HPARAMS = [
+    'magic',    # int32
+    'version',  # int32
+    'n_vocab',  # int32
+    'n_embd',   # int32
+    'n_mult',   # int32
+    'n_head',   # int32
+    'n_layer',  # int32
+    'n_rot',    # int32
+    'f16',      # int32
+]
+
+def read_hparams(fin):
+    struct_fmt = "i" * len(HPARAMS)
+    struct_size = struct.calcsize(struct_fmt)
+    buf = fin.read(struct_size)
+    ints = struct.unpack(struct_fmt, buf)
+    hparams = dict(zip(HPARAMS, ints))
+    return hparams
+
+def write_hparams(fout, hparams):
+    struct_fmt = "i" * len(HPARAMS)
+    struct_size = struct.calcsize(struct_fmt)
+    ints = [hparams[h] for h in HPARAMS]
+    fout.write(struct.pack(struct_fmt, *ints))
+
+def read_tokens(fin, hparams):
+    tokens = []
+    for i in range(hparams['n_vocab']):
+        len_b = fin.read(4)
+        (length,) = struct.unpack("i", len_b)
+        word = fin.read(length)
+        score_b = fin.read(4)
+        (score,) = struct.unpack("f", score_b)
+        tokens.append((word, score))
+    return tokens
+
+def write_tokens(fout, tokens):
+    for word, score in tokens:
+        fout.write(struct.pack("i", len(word)))
+        fout.write(word)
+        fout.write(struct.pack("f", score))
+
+def ggml_nelements(shape):
+    r = 1
+    for i in shape:
+        r *= i
+    return r
+
+def ggml_nbytes(shape, ftype):
+    x = ggml_nelements(shape)
+    t = WTYPES[ftype]
+    x *= GGML_TYPE_SIZE[t]
+    x //= GGML_BLCK_SIZE[t]
+    return x
+
+def copy_tensors(fin, fout, part_id, n_parts):
+    while True:
+
+        b = fin.read(4)
+        if not b: break
+        (n_dims,) = struct.unpack("i", b)
+        b = fin.read(4)
+        (length,) = struct.unpack("i", b)
+        b = fin.read(4)
+        (ftype,) = struct.unpack("i", b)
+
+        assert n_dims in (1, 2)
+
+        partshape = list(range(n_dims))
+        for i in range(n_dims):
+            b = fin.read(4)
+            partshape[i] = struct.unpack("i", b)[0]
+        partshape = list(reversed(partshape))
+
+        name = fin.read(length)
+        data = fin.read(ggml_nbytes(partshape, ftype))
+
+        blck_size = GGML_BLCK_SIZE[WTYPES[ftype]]
+        type_size = GGML_TYPE_SIZE[WTYPES[ftype]]
+
+        print(f"Processing tensor {name} with shape: {partshape} and type: {WTYPE_NAMES[ftype]}")
+
+        # determine dimension along which multipart tensor is sharded
+        #
+        # split_dim 0 regex:
+        #   - output.*
+        #   - layers.*.attention.wq.weight
+        #   - layers.*.attention.wk.weight
+        #   - layers.*.attention.wv.weight
+        #   - layers.*.feed_forward.w1.weight
+        #   - layers.*.feed_forward.w3.weight
+        #
+        # split_dim 1 regex:
+        #   - tok_embeddings.*
+        #   - layers.*.attention.wo.weight
+        #   - layers.*.feed_forward.w2.weight
+        #
+        if n_dims > 1:
+            split_dim = 1
+            if b"tok_embeddings" in name:
+                split_dim = 1
+            elif b"layers" in name:
+                if b"attention.wo.weight" in name:
+                    split_dim = 1
+                elif b"feed_forward.w2.weight" in name:
+                    split_dim = 1
+                else:
+                    split_dim = 0
+            elif b"output" in name:
+                split_dim = 0
+
+        # output tensor header
+        fullshape = list(partshape)
+        if n_dims > 1:
+            fullshape[split_dim] *= n_parts
+        fout.write(struct.pack("iii", n_dims, len(name), ftype))
+        for dim in reversed(fullshape):
+            fout.write(struct.pack("i", dim))
+        fout.write(name)
+
+        # ensure tensor data is aligned
+        tensor_data_offset = fout.tell()
+        while tensor_data_offset % QK != 0:
+            fout.write(struct.pack("B", 0))
+            tensor_data_offset += 1
+
+        # output unified mappable tensor data
+        if n_dims == 1 or n_parts == 1:
+            # copy tensor which we thankfully received in one piece
+            if part_id == 0:
+                fout.write(data)
+        elif split_dim == 0:
+            # reassemble multifile tensor containing some of the rows
+            rows_per_chunk = partshape[0]
+            current_row = part_id * rows_per_chunk
+            bytes_per_row = fullshape[1] // blck_size * type_size
+            offset = current_row * bytes_per_row
+            fout.seek(tensor_data_offset + offset)
+            fout.write(data)
+        elif split_dim == 1:
+            # reassemble multifile tensor containing some of the cols
+            cols_per_chunk = partshape[1]
+            current_col = part_id * cols_per_chunk
+            bpr = partshape[1] // blck_size * type_size
+            bytes_per_row = fullshape[1] // blck_size * type_size
+            offset_current_col = current_col // blck_size * type_size
+            for row in range(partshape[0]):
+                offset_row = row * bytes_per_row
+                offset = offset_row + offset_current_col
+                fout.seek(tensor_data_offset + offset)
+                fout.write(data[row * bpr:row * bpr + bpr])
+
+        # advance file position to next tensor
+        fout.seek(tensor_data_offset + ggml_nbytes(fullshape, ftype))
+
+def migrate(fin_path):
+    assert fin_path
+    assert os.path.exists(fin_path)
+
+    with open(fin_path, "rb") as fin:
+        hparams = read_hparams(fin)
+        tokens = read_tokens(fin, hparams)
+
+    if hparams['magic'] == 0x67676a74:  # ggjt
+        print("%s: input ggml has already been converted to 'ggjt' magic\n" %
+              (fin_path))
+        return
+
+    if hparams['magic'] != 0x67676d66:  # ggmf
+        print("%s: input ggml file doesn't have expected 'ggmf' magic: %#x\n" %
+              (fin_path, hparams['magic']))
+        return
+
+    hparams['magic'] = 0x67676a74  # ggjt
+
+    # count number of multipart files by convention
+    n_parts = 1
+    while True:
+        if os.path.exists("%s.%d" % (fin_path, n_parts)):
+            n_parts += 1
+        else:
+            break
+
+    # we output a single file for ggml
+    with open(fin_path+".migrated", "wb") as fout:
+        write_hparams(fout, hparams)
+        write_tokens(fout, tokens)
+        offset_of_tensors = fout.tell()
+        # the tensors we load could be split across multiple files
+        for part_id in range(n_parts):
+            fout.seek(offset_of_tensors)
+            print(f"Processing part {part_id+1} of {n_parts}\n")
+            fin_path = fin_path
+            if part_id > 0:
+                fin_path += ".%d" % (part_id)
+            with open(fin_path, "rb") as fin:
+                read_tokens(fin, read_hparams(fin))
+                copy_tensors(fin, fout, part_id, n_parts)
+
+    os.remove(fin_path)
+    os.rename(fin_path+".migrated", fin_path)
+
+    print(f"Done. Output file: {fin_path+'.migrated'}\n")