File size: 6,653 Bytes
08e5ef1
7edda8b
2bede7c
 
7edda8b
 
2bede7c
 
75b770e
08e5ef1
 
1fba392
 
925d15e
 
08e5ef1
2bede7c
69d19e7
925d15e
7686e09
 
 
 
 
 
 
 
 
 
2bede7c
d9267f6
7c36326
d9267f6
5696fee
 
f4651d4
9781999
d9267f6
75b770e
2124573
f4651d4
2124573
 
 
 
 
 
 
 
 
 
 
 
 
 
d9267f6
9781999
f4651d4
9781999
5696fee
9781999
b7ccecf
9781999
 
 
 
5696fee
9781999
 
 
 
 
 
 
2124573
5696fee
 
9781999
b7ccecf
d9267f6
b7ccecf
 
d9267f6
 
 
 
9781999
 
5696fee
ef80b76
9781999
 
 
ef80b76
b7ccecf
9781999
 
 
b7ccecf
9781999
b7ccecf
f4651d4
9781999
f4651d4
9781999
 
b7ccecf
f4651d4
9781999
5696fee
9781999
b7ccecf
2124573
b7ccecf
 
ef80b76
 
 
 
b7ccecf
9781999
 
5696fee
9781999
 
 
 
5696fee
9781999
 
 
5696fee
9781999
5696fee
9781999
 
 
 
5696fee
9781999
 
 
 
 
5696fee
9781999
 
2bede7c
 
 
f4651d4
2bede7c
1fba392
7edda8b
1fba392
 
7edda8b
 
f4651d4
 
7686e09
 
b416bb7
7edda8b
2124573
 
 
 
d9267f6
f4651d4
7cd57ad
 
 
 
87f5ccd
d9267f6
2bede7c
ec000c3
d2fb1de
ec000c3
 
2bede7c
925d15e
 
 
 
 
 
 
2bede7c
ec000c3
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
import os
import shutil
import subprocess

import gradio as gr

from huggingface_hub import create_repo, HfApi
from huggingface_hub import snapshot_download
from huggingface_hub import whoami
from huggingface_hub import ModelCard

from gradio_huggingfacehub_search import HuggingfaceHubSearch

from apscheduler.schedulers.background import BackgroundScheduler

from textwrap import dedent

LLAMA_LIKE_ARCHS = ["MistralForCausalLM",]
HF_TOKEN = os.environ.get("HF_TOKEN")

def script_to_use(model_id, api):
    info = api.model_info(model_id)
    if info.config is None:
        return None
    arch = info.config.get("architectures", None)
    if arch is None:
        return None
    arch = arch[0]
    return "convert.py" if arch in LLAMA_LIKE_ARCHS else "convert-hf-to-gguf.py"

def process_model(model_id, q_method, private_repo, oauth_token: gr.OAuthToken | None):
    if oauth_token.token is None:
        raise ValueError("You must be logged in to use GGUF-my-repo")
    model_name = model_id.split('/')[-1]
    fp16 = f"{model_name}/{model_name.lower()}.fp16.bin"

    try:
        api = HfApi(token=oauth_token.token)

        dl_pattern = ["*.md", "*.json", "*.model"]

        pattern = (
            "*.safetensors"
            if any(
                file.path.endswith(".safetensors")
                for file in api.list_repo_tree(
                    repo_id=model_id,
                    recursive=True,
                )
            )
            else "*.bin"
        )

        dl_pattern += pattern

        api.snapshot_download(repo_id=model_id, local_dir=model_name, local_dir_use_symlinks=False, allow_patterns=dl_pattern)
        print("Model downloaded successully!")

        conversion_script = script_to_use(model_id, api)
        fp16_conversion = f"python llama.cpp/{conversion_script} {model_name} --outtype f16 --outfile {fp16}"
        result = subprocess.run(fp16_conversion, shell=True, capture_output=True)
        print(result)
        if result.returncode != 0:
            raise Exception(f"Error converting to fp16: {result.stderr}")
        print("Model converted to fp16 successully!")

        qtype = f"{model_name}/{model_name.lower()}.{q_method.upper()}.gguf"
        quantise_ggml = f"./llama.cpp/quantize {fp16} {qtype} {q_method}"
        result = subprocess.run(quantise_ggml, shell=True, capture_output=True)
        if result.returncode != 0:
            raise Exception(f"Error quantizing: {result.stderr}")
        print("Quantised successfully!")

        # Create empty repo
        new_repo_url = api.create_repo(repo_id=f"{model_name}-{q_method}-GGUF", exist_ok=True, private=private_repo)
        new_repo_id = new_repo_url.repo_id
        print("Repo created successfully!", new_repo_url)

        try:
            card = ModelCard.load(model_id, token=oauth_token.token)
        except:
            card = ModelCard("")
        if card.data.tags is None:
            card.data.tags = []
        card.data.tags.append("llama-cpp")
        card.data.tags.append("gguf-my-repo")
        card.text = dedent(
            f"""
            # {new_repo_id}
            This model was converted to GGUF format from [`{model_id}`](https://huggingface.co/{model_id}) using llama.cpp via the ggml.ai's [GGUF-my-repo](https://huggingface.co/spaces/ggml-org/gguf-my-repo) space.
            Refer to the [original model card](https://huggingface.co/{model_id}) for more details on the model.
            ## Use with llama.cpp

            Install llama.cpp through brew.

            ```bash
            brew install ggerganov/ggerganov/llama.cpp
            ```
            Invoke the llama.cpp server or the CLI.

            CLI:

            ```bash
            llama-cli --hf-repo {new_repo_id} --model {qtype.split("/")[-1]} -p "The meaning to life and the universe is"
            ```

            Server:

            ```bash
            llama-server --hf-repo {new_repo_id} --model {qtype.split("/")[-1]} -c 2048
            ```

            Note: You can also use this checkpoint directly through the [usage steps](https://github.com/ggerganov/llama.cpp?tab=readme-ov-file#usage) listed in the Llama.cpp repo as well.

            ```
            git clone https://github.com/ggerganov/llama.cpp && \
            cd llama.cpp && \
            make && \
            ./main -m {qtype.split("/")[-1]} -n 128
            ```
            """
        )
        card.save(os.path.join(model_name, "README-new.md"))

        api.upload_file(
            path_or_fileobj=qtype,
            path_in_repo=qtype.split("/")[-1],
            repo_id=new_repo_id,
        )

        api.upload_file(
            path_or_fileobj=f"{model_name}/README-new.md",
            path_in_repo="README.md",
            repo_id=new_repo_id,
        )
        print("Uploaded successfully!")

        return (
            f'Find your repo <a href=\'{new_repo_url}\' target="_blank" style="text-decoration:underline">here</a>',
            "llama.png",
        )
    except Exception as e:
        return (f"Error: {e}", "error.png")
    finally:
        shutil.rmtree(model_name, ignore_errors=True)
        print("Folder cleaned up successfully!")


# Create Gradio interface
iface = gr.Interface(
    fn=process_model,
    inputs=[
        HuggingfaceHubSearch(
            label="Hub Model ID",
            placeholder="Search for model id on Huggingface",
            search_type="model",
        ),
        gr.Dropdown(
            ["Q2_K", "Q3_K_S", "Q3_K_M", "Q3_K_L", "Q4_0", "Q4_K_S", "Q4_K_M", "Q5_0", "Q5_K_S", "Q5_K_M", "Q6_K", "Q8_0"],
            label="Quantization Method",
            info="GGML quantisation type",
            value="Q4_K_M",
            filterable=False
        ),
        gr.Checkbox(
            value=False,
            label="Private Repo",
            info="Create a private repo under your username."
        ),
    ],
    outputs=[
        gr.Markdown(label="output"),
        gr.Image(show_label=False),
    ],
    title="Create your own GGUF Quants, blazingly fast ⚡!",
    description="The space takes an HF repo as an input, quantises it and creates a Public repo containing the selected quant under your HF user namespace.",
)
with gr.Blocks() as demo:
    gr.Markdown("You must be logged in to use GGUF-my-repo.")
    gr.LoginButton(min_width=250)
    iface.render()

def restart_space():
    HfApi().restart_space(repo_id="ggml-org/gguf-my-repo", token=HF_TOKEN)

scheduler = BackgroundScheduler()
scheduler.add_job(restart_space, "interval", seconds=86400)
scheduler.start()

# Launch the interface
demo.queue(default_concurrency_limit=1, max_size=5).launch(debug=True)