Merge pull request #7464 from oobabooga/dev

Merge dev branch

.github/workflows/build-everything-tgw.yml

@@ -41,6 +41,13 @@ jobs:
       version: ${{ inputs.version }}
       config: 'os:ubuntu-22.04'
 
+  build_release_rocm_windows:
+    name: ROCm Windows
+    uses: ./.github/workflows/build-portable-release-rocm.yml
+    with:
+      version: ${{ inputs.version }}
+      config: 'os:windows-2022'
+
   build_release_rocm_linux:
     name: ROCm Linux
     uses: ./.github/workflows/build-portable-release-rocm.yml

modules/chat.py

@@ -210,6 +210,57 @@ def _expand_tool_sequence(tool_seq):
     return messages
 
 
+def _convert_to_tool_responses(messages):
+    """Convert role:'tool' messages to tool_responses format.
+
+    Templates like Gemma 4 expect tool results as a ``tool_responses``
+    attribute on a message rather than separate ``role: 'tool'`` messages.
+    This function groups consecutive tool messages and rewrites them.
+    """
+    result = []
+    tc_id_to_name = {}
+
+    i = 0
+    while i < len(messages):
+        msg = messages[i]
+
+        if msg.get('tool_calls'):
+            for tc in msg['tool_calls']:
+                tc_id = tc.get('id', '')
+                func_name = tc.get('function', {}).get('name', 'unknown')
+                if tc_id:
+                    tc_id_to_name[tc_id] = func_name
+
+        if msg.get('role') == 'tool':
+            tool_responses = []
+            while i < len(messages) and messages[i].get('role') == 'tool':
+                tool_msg = messages[i]
+                tc_id = tool_msg.get('tool_call_id', '')
+                func_name = tc_id_to_name.get(tc_id, 'unknown')
+
+                content = tool_msg.get('content', '')
+                try:
+                    response = json.loads(content)
+                except (json.JSONDecodeError, ValueError, TypeError):
+                    response = content
+
+                tool_responses.append({
+                    'name': func_name,
+                    'response': response,
+                })
+                i += 1
+
+            result.append({
+                'role': 'tool',
+                'tool_responses': tool_responses,
+            })
+        else:
+            result.append(msg)
+            i += 1
+
+    return result
+
+
 def _format_attachments(attachments, include_text=True):
     """Build image ref and text attachment strings from a list of attachments."""
     attachments_text = ""
@@ -267,6 +318,9 @@ def generate_chat_prompt(user_input, state, **kwargs):
         tools=state['tools'] if 'tools' in state else None,
     )
 
+    active_template_str = state['instruction_template_str'] if state['mode'] == 'instruct' else chat_template_str
+    uses_tool_responses = 'tool_responses' in active_template_str
+
     messages = []
 
     if state['mode'] == 'instruct':
@@ -503,6 +557,9 @@ def generate_chat_prompt(user_input, state, **kwargs):
 
         return prompt
 
+    if uses_tool_responses:
+        messages = _convert_to_tool_responses(messages)
+
     prompt = make_prompt(messages)
 
     # Handle truncation

modules/prompts.py

@@ -1,6 +1,7 @@
 from pathlib import Path
 
 from modules import shared, utils
+from modules.utils import sanitize_filename
 from modules.text_generation import get_encoded_length
 
 
@@ -18,6 +19,7 @@ def load_prompt(fname):
 
         return initial_content
 
+    fname = sanitize_filename(fname)
     file_path = shared.user_data_dir / 'logs' / 'notebook' / f'{fname}.txt'
     if file_path.exists():
         with open(file_path, 'r', encoding='utf-8') as f:

modules/reasoning.py

@@ -7,6 +7,7 @@ THINKING_FORMATS = [
     ('<|channel|>analysis<|message|>', '<|end|>', '<|channel|>final<|message|>'),
     ('<|channel|>commentary<|message|>', '<|end|>', '<|channel|>final<|message|>'),
     ('<seed:think>', '</seed:think>', None),
+    ('<|channel>thought', '<channel|>', None),  # Gemma 4
     ('<|think|>', '<|end|>', '<|content|>'),  # Solar Open
     # ('Thinking Process:', '</think>', None),  # Qwen3.5 verbose thinking outside tags -- removed: too prone to false positives in streaming
     (None, '</think>', None),  # End-only variant (e.g., Qwen3-next)

modules/tool_parsing.py

@@ -27,6 +27,7 @@ TOOL_CALL_OPENING_MARKERS = [
     '[TOOL_CALLS]',
     'to=functions.',
     '<|channel|>commentary',
+    '<|tool_call>call:',
 ]
 
 
@@ -400,6 +401,78 @@ def _parse_glm_tool_calls(answer: str, tool_names: list[str]):
     return matches, start_pos
 
 
+def _extract_gemma4_balanced(text, start):
+    """Extract balanced braces from Gemma 4 format, using <|"|> as string delimiters."""
+    if start >= len(text) or text[start] != '{':
+        return None
+    depth = 0
+    in_string = False
+    quote_token = '<|"|>'
+    quote_len = len(quote_token)
+    i = start
+    while i < len(text):
+        if text[i:i + quote_len] == quote_token:
+            in_string = not in_string
+            i += quote_len
+            continue
+        if in_string:
+            i += 1
+            continue
+        c = text[i]
+        if c == '{':
+            depth += 1
+        elif c == '}':
+            depth -= 1
+            if depth == 0:
+                return text[start:i + 1]
+        i += 1
+    return None
+
+
+def _parse_gemma4_tool_calls(answer: str, tool_names: list[str]):
+    """Parse Gemma 4-style tool calls.
+
+    Format:
+        <|tool_call>call:func_name{key:<|"|>value<|"|>,...}<tool_call|>
+
+    Values use <|"|> tokens instead of standard JSON quotes, and keys are
+    bare identifiers.
+    """
+    matches = []
+    start_pos = None
+
+    for m in re.finditer(r'<\|tool_call>call:([^\s{]+)\s*', answer):
+        func_name = m.group(1).strip()
+        if func_name not in tool_names:
+            continue
+
+        brace_start = m.end()
+        if brace_start >= len(answer) or answer[brace_start] != '{':
+            continue
+
+        content = _extract_gemma4_balanced(answer, brace_start)
+        if content is None:
+            continue
+
+        # Convert to JSON: split on <|"|> tokens so that key quoting
+        # only applies outside string values (even-indexed parts),
+        # then rejoin with real quotes.
+        parts = content.split('<|"|>')
+        for idx in range(0, len(parts), 2):
+            parts[idx] = re.sub(r'(^|[{,\[])\s*(\w+)\s*:', r'\1"\2":', parts[idx])
+        json_str = '"'.join(parts)
+
+        try:
+            arguments = json.loads(json_str)
+            if start_pos is None:
+                start_pos = m.start()
+            matches.append(_make_tool_call(func_name, arguments))
+        except (json.JSONDecodeError, ValueError):
+            pass
+
+    return matches, start_pos
+
+
 def _parse_pythonic_tool_calls(answer: str, tool_names: list[str]):
     """Parse pythonic-style tool calls used by Llama 4 and similar models.
 
@@ -472,6 +545,11 @@ TOOL_CALL_FORMATS = [
         'parser': _parse_channel_tool_calls,
         'markers': ['to=functions.', '<|channel|>commentary'],
     },
+    {
+        'template_hints': ['<|tool_call>call:'],
+        'parser': _parse_gemma4_tool_calls,
+        'markers': ['<|tool_call>call:'],
+    },
     {
         'template_hints': ['minimax:tool_call'],
         'parser': _parse_minimax_tool_calls,
@@ -504,6 +582,7 @@ ALL_PARSERS = [
     _parse_deep_seek_tool_calls,
     _parse_kimi_tool_calls,
     _parse_channel_tool_calls,
+    _parse_gemma4_tool_calls,
     _parse_minimax_tool_calls,
     _parse_glm_tool_calls,
     _parse_xml_param_tool_calls,
@@ -620,6 +699,8 @@ def parse_tool_call(answer: str, tool_names: list[str], return_prefix: bool = Fa
                 if not isinstance(candidates, list):
                     candidates = [candidates]
                 for candidate_dict in candidates:
+                    if not isinstance(candidate_dict, dict):
+                        continue
                     checked_candidate = check_and_sanitize_tool_call_candidate(candidate_dict, tool_names)
                     if checked_candidate is not None:
                         matches.append(checked_candidate)

modules/ui_default.py

@@ -10,7 +10,7 @@ from modules.text_generation import (
     stop_everything_event
 )
 from modules.ui_notebook import store_notebook_state_and_debounce
-from modules.utils import gradio
+from modules.utils import gradio, sanitize_filename
 
 inputs = ('textbox-default', 'interface_state')
 outputs = ('output_textbox', 'html-default')
@@ -167,6 +167,7 @@ def handle_new_prompt():
 
 
 def handle_delete_prompt_confirm_default(prompt_name):
+    prompt_name = sanitize_filename(prompt_name)
     available_prompts = utils.get_available_prompts()
     current_index = available_prompts.index(prompt_name) if prompt_name in available_prompts else 0
 
@@ -199,6 +200,8 @@ def handle_rename_prompt_click_default(current_name):
 
 
 def handle_rename_prompt_confirm_default(new_name, current_name):
+    new_name = sanitize_filename(new_name)
+    current_name = sanitize_filename(current_name)
     old_path = shared.user_data_dir / "logs" / "notebook" / f"{current_name}.txt"
     new_path = shared.user_data_dir / "logs" / "notebook" / f"{new_name}.txt"
 

modules/ui_notebook.py

@@ -11,7 +11,7 @@ from modules.text_generation import (
     get_token_ids,
     stop_everything_event
 )
-from modules.utils import gradio
+from modules.utils import gradio, sanitize_filename
 
 _notebook_file_lock = threading.Lock()
 _notebook_auto_save_timer = None
@@ -202,6 +202,7 @@ def handle_new_prompt():
 
 
 def handle_delete_prompt_confirm_notebook(prompt_name):
+    prompt_name = sanitize_filename(prompt_name)
     available_prompts = utils.get_available_prompts()
     current_index = available_prompts.index(prompt_name) if prompt_name in available_prompts else 0
 
@@ -233,6 +234,8 @@ def handle_rename_prompt_click_notebook(current_name):
 
 
 def handle_rename_prompt_confirm_notebook(new_name, current_name):
+    new_name = sanitize_filename(new_name)
+    current_name = sanitize_filename(current_name)
     old_path = shared.user_data_dir / "logs" / "notebook" / f"{current_name}.txt"
     new_path = shared.user_data_dir / "logs" / "notebook" / f"{new_name}.txt"
 
@@ -249,6 +252,7 @@ def handle_rename_prompt_confirm_notebook(new_name, current_name):
 
 def autosave_prompt(text, prompt_name):
     """Automatically save the text to the selected prompt file"""
+    prompt_name = sanitize_filename(prompt_name)
     if prompt_name and text.strip():
         prompt_path = shared.user_data_dir / "logs" / "notebook" / f"{prompt_name}.txt"
         prompt_path.parent.mkdir(parents=True, exist_ok=True)

modules/utils.py

@@ -105,6 +105,9 @@ def resolve_model_path(model_name_or_path, image_model=False):
     before the default models directory.
     """
 
+    if model_name_or_path is None:
+        raise FileNotFoundError("No model specified.")
+
     path_candidate = Path(model_name_or_path)
     if path_candidate.exists():
         return path_candidate

requirements/full/requirements.txt

@@ -31,8 +31,8 @@ tqdm
 wandb
 
 # Gradio
-https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.14/gradio-4.37.2+custom.14-py3-none-any.whl
+https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.18/gradio-4.37.2+custom.18-py3-none-any.whl
-https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.14/gradio_client-1.0.2+custom.14-py3-none-any.whl
+https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.18/gradio_client-1.0.2+custom.18-py3-none-any.whl
 
 # API
 flask_cloudflared==0.0.15
@@ -40,10 +40,10 @@ sse-starlette==1.6.5
 tiktoken
 
 # CUDA wheels
-https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.102.0/llama_cpp_binaries-0.102.0+cu124-py3-none-win_amd64.whl; platform_system == "Windows"
+https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.106.0/llama_cpp_binaries-0.106.0+cu124-py3-none-win_amd64.whl; platform_system == "Windows"
-https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.102.0/llama_cpp_binaries-0.102.0+cu124-py3-none-linux_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64"
+https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.106.0/llama_cpp_binaries-0.106.0+cu124-py3-none-linux_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64"
-https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.102.0/ik_llama_cpp_binaries-0.102.0+cu124-py3-none-win_amd64.whl; platform_system == "Windows"
+https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.106.0/ik_llama_cpp_binaries-0.106.0+cu124-py3-none-win_amd64.whl; platform_system == "Windows"
-https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.102.0/ik_llama_cpp_binaries-0.102.0+cu124-py3-none-linux_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64"
+https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.106.0/ik_llama_cpp_binaries-0.106.0+cu124-py3-none-linux_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64"
 https://github.com/turboderp-org/exllamav3/releases/download/v0.0.28/exllamav3-0.0.28+cu128.torch2.9.0-cp313-cp313-win_amd64.whl; platform_system == "Windows" and python_version == "3.13"
 https://github.com/turboderp-org/exllamav3/releases/download/v0.0.28/exllamav3-0.0.28+cu128.torch2.9.0-cp313-cp313-linux_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.13"
 https://github.com/kingbri1/flash-attention/releases/download/v2.8.3/flash_attn-2.8.3+cu128torch2.9.0cxx11abiFALSE-cp313-cp313-win_amd64.whl; platform_system == "Windows" and python_version == "3.13"

requirements/full/requirements_amd.txt

@@ -28,8 +28,8 @@ trafilatura==2.0.0
 wandb
 
 # Gradio
-https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.14/gradio-4.37.2+custom.14-py3-none-any.whl
+https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.18/gradio-4.37.2+custom.18-py3-none-any.whl
-https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.14/gradio_client-1.0.2+custom.14-py3-none-any.whl
+https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.18/gradio_client-1.0.2+custom.18-py3-none-any.whl
 
 # API
 flask_cloudflared==0.0.15
@@ -37,5 +37,5 @@ sse-starlette==1.6.5
 tiktoken
 
 # AMD wheels
-https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.102.0/llama_cpp_binaries-0.102.0+rocm7.2-py3-none-win_amd64.whl; platform_system == "Windows"
+https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.106.0/llama_cpp_binaries-0.106.0+rocm7.2-py3-none-win_amd64.whl; platform_system == "Windows"
-https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.102.0/llama_cpp_binaries-0.102.0+rocm7.2-py3-none-linux_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64"
+https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.106.0/llama_cpp_binaries-0.106.0+rocm7.2-py3-none-linux_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64"

requirements/full/requirements_apple_intel.txt

@@ -28,8 +28,8 @@ trafilatura==2.0.0
 wandb
 
 # Gradio
-https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.14/gradio-4.37.2+custom.14-py3-none-any.whl
+https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.18/gradio-4.37.2+custom.18-py3-none-any.whl
-https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.14/gradio_client-1.0.2+custom.14-py3-none-any.whl
+https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.18/gradio_client-1.0.2+custom.18-py3-none-any.whl
 
 # API
 flask_cloudflared==0.0.15
@@ -37,4 +37,4 @@ sse-starlette==1.6.5
 tiktoken
 
 # Mac wheels
-https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.102.0/llama_cpp_binaries-0.102.0-py3-none-macosx_13_0_x86_64.whl; platform_system == "Darwin"
+https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.106.0/llama_cpp_binaries-0.106.0-py3-none-macosx_13_0_x86_64.whl; platform_system == "Darwin"

requirements/full/requirements_apple_silicon.txt

@@ -28,8 +28,8 @@ trafilatura==2.0.0
 wandb
 
 # Gradio
-https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.14/gradio-4.37.2+custom.14-py3-none-any.whl
+https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.18/gradio-4.37.2+custom.18-py3-none-any.whl
-https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.14/gradio_client-1.0.2+custom.14-py3-none-any.whl
+https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.18/gradio_client-1.0.2+custom.18-py3-none-any.whl
 
 # API
 flask_cloudflared==0.0.15
@@ -37,4 +37,4 @@ sse-starlette==1.6.5
 tiktoken
 
 # Mac wheels
-https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.102.0/llama_cpp_binaries-0.102.0-py3-none-macosx_13_0_arm64.whl; platform_system == "Darwin"
+https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.106.0/llama_cpp_binaries-0.106.0-py3-none-macosx_13_0_arm64.whl; platform_system == "Darwin"

requirements/full/requirements_cpu_only.txt

@@ -28,8 +28,8 @@ trafilatura==2.0.0
 wandb
 
 # Gradio
-https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.14/gradio-4.37.2+custom.14-py3-none-any.whl
+https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.18/gradio-4.37.2+custom.18-py3-none-any.whl
-https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.14/gradio_client-1.0.2+custom.14-py3-none-any.whl
+https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.18/gradio_client-1.0.2+custom.18-py3-none-any.whl
 
 # API
 flask_cloudflared==0.0.15
@@ -37,7 +37,7 @@ sse-starlette==1.6.5
 tiktoken
 
 # llama.cpp (CPU only)
-https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.102.0/llama_cpp_binaries-0.102.0+cpu-py3-none-linux_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64"
+https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.106.0/llama_cpp_binaries-0.106.0+cpu-py3-none-linux_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64"
-https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.102.0/llama_cpp_binaries-0.102.0+cpu-py3-none-win_amd64.whl; platform_system == "Windows"
+https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.106.0/llama_cpp_binaries-0.106.0+cpu-py3-none-win_amd64.whl; platform_system == "Windows"
-https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.102.0/ik_llama_cpp_binaries-0.102.0+cpu-py3-none-linux_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64"
+https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.106.0/ik_llama_cpp_binaries-0.106.0+cpu-py3-none-linux_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64"
-https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.102.0/ik_llama_cpp_binaries-0.102.0+cpu-py3-none-win_amd64.whl; platform_system == "Windows"
+https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.106.0/ik_llama_cpp_binaries-0.106.0+cpu-py3-none-win_amd64.whl; platform_system == "Windows"

requirements/full/requirements_nowheels.txt

@@ -28,8 +28,8 @@ trafilatura==2.0.0
 wandb
 
 # Gradio
-https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.14/gradio-4.37.2+custom.14-py3-none-any.whl
+https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.18/gradio-4.37.2+custom.18-py3-none-any.whl
-https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.14/gradio_client-1.0.2+custom.14-py3-none-any.whl
+https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.18/gradio_client-1.0.2+custom.18-py3-none-any.whl
 
 # API
 flask_cloudflared==0.0.15

requirements/portable/requirements.txt

@@ -14,8 +14,8 @@ trafilatura==2.0.0
 tqdm
 
 # Gradio
-https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.14/gradio-4.37.2+custom.14-py3-none-any.whl
+https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.18/gradio-4.37.2+custom.18-py3-none-any.whl
-https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.14/gradio_client-1.0.2+custom.14-py3-none-any.whl
+https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.18/gradio_client-1.0.2+custom.18-py3-none-any.whl
 
 # API
 flask_cloudflared==0.0.15
@@ -23,5 +23,5 @@ sse-starlette==1.6.5
 tiktoken
 
 # CUDA wheels
-https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.102.0/llama_cpp_binaries-0.102.0+cu124-py3-none-win_amd64.whl; platform_system == "Windows"
+https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.106.0/llama_cpp_binaries-0.106.0+cu124-py3-none-win_amd64.whl; platform_system == "Windows"
-https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.102.0/llama_cpp_binaries-0.102.0+cu124-py3-none-linux_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64"
+https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.106.0/llama_cpp_binaries-0.106.0+cu124-py3-none-linux_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64"

requirements/portable/requirements_amd.txt

@@ -14,8 +14,8 @@ trafilatura==2.0.0
 tqdm
 
 # Gradio
-https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.14/gradio-4.37.2+custom.14-py3-none-any.whl
+https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.18/gradio-4.37.2+custom.18-py3-none-any.whl
-https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.14/gradio_client-1.0.2+custom.14-py3-none-any.whl
+https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.18/gradio_client-1.0.2+custom.18-py3-none-any.whl
 
 # API
 flask_cloudflared==0.0.15
@@ -23,5 +23,5 @@ sse-starlette==1.6.5
 tiktoken
 
 # AMD wheels
-https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.102.0/llama_cpp_binaries-0.102.0+rocm7.2-py3-none-win_amd64.whl; platform_system == "Windows"
+https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.106.0/llama_cpp_binaries-0.106.0+rocm7.2-py3-none-win_amd64.whl; platform_system == "Windows"
-https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.102.0/llama_cpp_binaries-0.102.0+rocm7.2-py3-none-linux_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64"
+https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.106.0/llama_cpp_binaries-0.106.0+rocm7.2-py3-none-linux_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64"

requirements/portable/requirements_apple_intel.txt

@@ -14,8 +14,8 @@ trafilatura==2.0.0
 tqdm
 
 # Gradio
-https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.14/gradio-4.37.2+custom.14-py3-none-any.whl
+https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.18/gradio-4.37.2+custom.18-py3-none-any.whl
-https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.14/gradio_client-1.0.2+custom.14-py3-none-any.whl
+https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.18/gradio_client-1.0.2+custom.18-py3-none-any.whl
 
 # API
 flask_cloudflared==0.0.15
@@ -23,4 +23,4 @@ sse-starlette==1.6.5
 tiktoken
 
 # Mac wheels
-https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.102.0/llama_cpp_binaries-0.102.0-py3-none-macosx_13_0_x86_64.whl; platform_system == "Darwin"
+https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.106.0/llama_cpp_binaries-0.106.0-py3-none-macosx_13_0_x86_64.whl; platform_system == "Darwin"

requirements/portable/requirements_apple_silicon.txt

@@ -14,8 +14,8 @@ trafilatura==2.0.0
 tqdm
 
 # Gradio
-https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.14/gradio-4.37.2+custom.14-py3-none-any.whl
+https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.18/gradio-4.37.2+custom.18-py3-none-any.whl
-https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.14/gradio_client-1.0.2+custom.14-py3-none-any.whl
+https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.18/gradio_client-1.0.2+custom.18-py3-none-any.whl
 
 # API
 flask_cloudflared==0.0.15
@@ -23,4 +23,4 @@ sse-starlette==1.6.5
 tiktoken
 
 # Mac wheels
-https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.102.0/llama_cpp_binaries-0.102.0-py3-none-macosx_13_0_arm64.whl; platform_system == "Darwin"
+https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.106.0/llama_cpp_binaries-0.106.0-py3-none-macosx_13_0_arm64.whl; platform_system == "Darwin"

requirements/portable/requirements_cpu_only.txt

@@ -14,8 +14,8 @@ trafilatura==2.0.0
 tqdm
 
 # Gradio
-https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.14/gradio-4.37.2+custom.14-py3-none-any.whl
+https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.18/gradio-4.37.2+custom.18-py3-none-any.whl
-https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.14/gradio_client-1.0.2+custom.14-py3-none-any.whl
+https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.18/gradio_client-1.0.2+custom.18-py3-none-any.whl
 
 # API
 flask_cloudflared==0.0.15
@@ -23,5 +23,5 @@ sse-starlette==1.6.5
 tiktoken
 
 # llama.cpp (CPU only)
-https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.102.0/llama_cpp_binaries-0.102.0+cpu-py3-none-linux_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64"
+https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.106.0/llama_cpp_binaries-0.106.0+cpu-py3-none-linux_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64"
-https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.102.0/llama_cpp_binaries-0.102.0+cpu-py3-none-win_amd64.whl; platform_system == "Windows"
+https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.106.0/llama_cpp_binaries-0.106.0+cpu-py3-none-win_amd64.whl; platform_system == "Windows"

requirements/portable/requirements_cuda131.txt

@@ -14,8 +14,8 @@ trafilatura==2.0.0
 tqdm
 
 # Gradio
-https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.14/gradio-4.37.2+custom.14-py3-none-any.whl
+https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.18/gradio-4.37.2+custom.18-py3-none-any.whl
-https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.14/gradio_client-1.0.2+custom.14-py3-none-any.whl
+https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.18/gradio_client-1.0.2+custom.18-py3-none-any.whl
 
 # API
 flask_cloudflared==0.0.15
@@ -23,5 +23,5 @@ sse-starlette==1.6.5
 tiktoken
 
 # CUDA wheels
-https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.102.0/llama_cpp_binaries-0.102.0+cu131-py3-none-win_amd64.whl; platform_system == "Windows"
+https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.106.0/llama_cpp_binaries-0.106.0+cu131-py3-none-win_amd64.whl; platform_system == "Windows"
-https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.102.0/llama_cpp_binaries-0.102.0+cu131-py3-none-linux_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64"
+https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.106.0/llama_cpp_binaries-0.106.0+cu131-py3-none-linux_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64"

requirements/portable/requirements_ik.txt

@@ -14,8 +14,8 @@ trafilatura==2.0.0
 tqdm
 
 # Gradio
-https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.14/gradio-4.37.2+custom.14-py3-none-any.whl
+https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.18/gradio-4.37.2+custom.18-py3-none-any.whl
-https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.14/gradio_client-1.0.2+custom.14-py3-none-any.whl
+https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.18/gradio_client-1.0.2+custom.18-py3-none-any.whl
 
 # API
 flask_cloudflared==0.0.15
@@ -23,5 +23,5 @@ sse-starlette==1.6.5
 tiktoken
 
 # CUDA wheels
-https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.102.0/ik_llama_cpp_binaries-0.102.0+cu124-py3-none-win_amd64.whl; platform_system == "Windows"
+https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.106.0/ik_llama_cpp_binaries-0.106.0+cu124-py3-none-win_amd64.whl; platform_system == "Windows"
-https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.102.0/ik_llama_cpp_binaries-0.102.0+cu124-py3-none-linux_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64"
+https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.106.0/ik_llama_cpp_binaries-0.106.0+cu124-py3-none-linux_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64"

requirements/portable/requirements_ik_cpu_only.txt

@@ -14,8 +14,8 @@ trafilatura==2.0.0
 tqdm
 
 # Gradio
-https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.14/gradio-4.37.2+custom.14-py3-none-any.whl
+https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.18/gradio-4.37.2+custom.18-py3-none-any.whl
-https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.14/gradio_client-1.0.2+custom.14-py3-none-any.whl
+https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.18/gradio_client-1.0.2+custom.18-py3-none-any.whl
 
 # API
 flask_cloudflared==0.0.15
@@ -23,5 +23,5 @@ sse-starlette==1.6.5
 tiktoken
 
 # ik_llama.cpp (CPU only)
-https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.102.0/ik_llama_cpp_binaries-0.102.0+cpu-py3-none-linux_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64"
+https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.106.0/ik_llama_cpp_binaries-0.106.0+cpu-py3-none-linux_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64"
-https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.102.0/ik_llama_cpp_binaries-0.102.0+cpu-py3-none-win_amd64.whl; platform_system == "Windows"
+https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.106.0/ik_llama_cpp_binaries-0.106.0+cpu-py3-none-win_amd64.whl; platform_system == "Windows"

requirements/portable/requirements_ik_cuda131.txt

@@ -14,8 +14,8 @@ trafilatura==2.0.0
 tqdm
 
 # Gradio
-https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.14/gradio-4.37.2+custom.14-py3-none-any.whl
+https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.18/gradio-4.37.2+custom.18-py3-none-any.whl
-https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.14/gradio_client-1.0.2+custom.14-py3-none-any.whl
+https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.18/gradio_client-1.0.2+custom.18-py3-none-any.whl
 
 # API
 flask_cloudflared==0.0.15
@@ -23,5 +23,5 @@ sse-starlette==1.6.5
 tiktoken
 
 # CUDA wheels
-https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.102.0/ik_llama_cpp_binaries-0.102.0+cu131-py3-none-win_amd64.whl; platform_system == "Windows"
+https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.106.0/ik_llama_cpp_binaries-0.106.0+cu131-py3-none-win_amd64.whl; platform_system == "Windows"
-https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.102.0/ik_llama_cpp_binaries-0.102.0+cu131-py3-none-linux_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64"
+https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.106.0/ik_llama_cpp_binaries-0.106.0+cu131-py3-none-linux_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64"

requirements/portable/requirements_nowheels.txt

@@ -14,8 +14,8 @@ trafilatura==2.0.0
 tqdm
 
 # Gradio
-https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.14/gradio-4.37.2+custom.14-py3-none-any.whl
+https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.18/gradio-4.37.2+custom.18-py3-none-any.whl
-https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.14/gradio_client-1.0.2+custom.14-py3-none-any.whl
+https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.18/gradio_client-1.0.2+custom.18-py3-none-any.whl
 
 # API
 flask_cloudflared==0.0.15

requirements/portable/requirements_vulkan.txt

@@ -14,8 +14,8 @@ trafilatura==2.0.0
 tqdm
 
 # Gradio
-https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.14/gradio-4.37.2+custom.14-py3-none-any.whl
+https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.18/gradio-4.37.2+custom.18-py3-none-any.whl
-https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.14/gradio_client-1.0.2+custom.14-py3-none-any.whl
+https://github.com/oobabooga/gradio/releases/download/4.37.2-custom.18/gradio_client-1.0.2+custom.18-py3-none-any.whl
 
 # API
 flask_cloudflared==0.0.15
@@ -23,5 +23,5 @@ sse-starlette==1.6.5
 tiktoken
 
 # Vulkan wheels
-https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.102.0/llama_cpp_binaries-0.102.0+vulkan-py3-none-win_amd64.whl; platform_system == "Windows"
+https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.106.0/llama_cpp_binaries-0.106.0+vulkan-py3-none-win_amd64.whl; platform_system == "Windows"
-https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.102.0/llama_cpp_binaries-0.102.0+vulkan-py3-none-linux_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64"
+https://github.com/oobabooga/llama-cpp-binaries/releases/download/v0.106.0/llama_cpp_binaries-0.106.0+vulkan-py3-none-linux_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64"