python-fastllm 0.0.30__tar.gz → 0.0.32__tar.gz

{python_fastllm-0.0.30 → python_fastllm-0.0.32}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: python-fastllm
-Version: 0.0.30
+Version: 0.0.32
 Author-email: Kerem Turgutlu <keremturgutlu@gmail.com>
 License: Apache-2.0
 Project-URL: Repository, https://github.com/AnswerDotAI/fastllm
@@ -9,6 +9,8 @@ Classifier: Programming Language :: Python :: 3
 Classifier: Programming Language :: Python :: 3 :: Only
 Requires-Python: >=3.10
 Description-Content-Type: text/markdown
+Requires-Dist: fastcore>=2.1.11
+Requires-Dist: aidialog>=0.0.3
 Requires-Dist: fastspec>=0.0.11
 Requires-Dist: toolslm
 Requires-Dist: pillow

python_fastllm-0.0.32/fastllm/__init__.py

@@ -0,0 +1 @@
+__version__ = "0.0.32"

{python_fastllm-0.0.30 → python_fastllm-0.0.32}/fastllm/_modidx.py

@@ -16,7 +16,8 @@ d = { 'settings': { 'branch': 'main',
                                    'fastllm.acomplete._retry_call': ('acomplete.html#_retry_call', 'fastllm/acomplete.py'),
                                    'fastllm.acomplete._retry_stream': ('acomplete.html#_retry_stream', 'fastllm/acomplete.py'),
                                    'fastllm.acomplete.acomplete': ('acomplete.html#acomplete', 'fastllm/acomplete.py'),
-                                   'fastllm.acomplete.mk_client': ('acomplete.html#mk_client', 'fastllm/acomplete.py')},
+                                   'fastllm.acomplete.mk_client': ('acomplete.html#mk_client', 'fastllm/acomplete.py'),
+                                   'fastllm.acomplete.split_vendor': ('acomplete.html#split_vendor', 'fastllm/acomplete.py')},
             'fastllm.anthropic': { 'fastllm.anthropic._ant_cc': ('anthropic.html#_ant_cc', 'fastllm/anthropic.py'),
                                    'fastllm.anthropic._ant_part_type': ('anthropic.html#_ant_part_type', 'fastllm/anthropic.py'),
                                    'fastllm.anthropic._sanid': ('anthropic.html#_sanid', 'fastllm/anthropic.py'),
@@ -77,14 +78,9 @@ d = { 'settings': { 'branch': 'main',
                               'fastllm.chat.FenceToolStop': ('chat.html#fencetoolstop', 'fastllm/chat.py'),
                               'fastllm.chat.FenceToolStop.__call__': ('chat.html#fencetoolstop.__call__', 'fastllm/chat.py'),
                               'fastllm.chat.FenceToolStop.__init__': ('chat.html#fencetoolstop.__init__', 'fastllm/chat.py'),
-                              'fastllm.chat.FullResponse': ('chat.html#fullresponse', 'fastllm/chat.py'),
-                              'fastllm.chat.MediaUrl': ('chat.html#mediaurl', 'fastllm/chat.py'),
-                              'fastllm.chat.MediaUrl.__init__': ('chat.html#mediaurl.__init__', 'fastllm/chat.py'),
-                              'fastllm.chat.Msg.text': ('chat.html#msg.text', 'fastllm/chat.py'),
                               'fastllm.chat.StopReasonCallback': ('chat.html#stopreasoncallback', 'fastllm/chat.py'),
                               'fastllm.chat.StopReasonCallback.after_acomplete': ( 'chat.html#stopreasoncallback.after_acomplete',
                                                                                    'fastllm/chat.py'),
-                              'fastllm.chat.StopResponse': ('chat.html#stopresponse', 'fastllm/chat.py'),
                               'fastllm.chat.StopSequencesCallback': ('chat.html#stopsequencescallback', 'fastllm/chat.py'),
                               'fastllm.chat.StopSequencesCallback.__init__': ( 'chat.html#stopsequencescallback.__init__',
                                                                                'fastllm/chat.py'),
@@ -112,7 +108,6 @@ d = { 'settings': { 'branch': 'main',
                               'fastllm.chat._apply_cache_idxs': ('chat.html#_apply_cache_idxs', 'fastllm/chat.py'),
                               'fastllm.chat._bytes2content': ('chat.html#_bytes2content', 'fastllm/chat.py'),
                               'fastllm.chat._call_func': ('chat.html#_call_func', 'fastllm/chat.py'),
-                              'fastllm.chat._extract_tool_parts': ('chat.html#_extract_tool_parts', 'fastllm/chat.py'),
                               'fastllm.chat._handle_stop_reason': ('chat.html#_handle_stop_reason', 'fastllm/chat.py'),
                               'fastllm.chat._has_cache': ('chat.html#_has_cache', 'fastllm/chat.py'),
                               'fastllm.chat._has_search': ('chat.html#_has_search', 'fastllm/chat.py'),
@@ -122,15 +117,10 @@ d = { 'settings': { 'branch': 'main',
                               'fastllm.chat._mime2part_type': ('chat.html#_mime2part_type', 'fastllm/chat.py'),
                               'fastllm.chat._mk_content': ('chat.html#_mk_content', 'fastllm/chat.py'),
                               'fastllm.chat._mk_prefill': ('chat.html#_mk_prefill', 'fastllm/chat.py'),
-                              'fastllm.chat._mk_result_fence': ('chat.html#_mk_result_fence', 'fastllm/chat.py'),
                               'fastllm.chat._mk_tool_result': ('chat.html#_mk_tool_result', 'fastllm/chat.py'),
-                              'fastllm.chat._split_fence_msgs': ('chat.html#_split_fence_msgs', 'fastllm/chat.py'),
-                              'fastllm.chat._split_msg_on_fences': ('chat.html#_split_msg_on_fences', 'fastllm/chat.py'),
                               'fastllm.chat._tc_summary': ('chat.html#_tc_summary', 'fastllm/chat.py'),
                               'fastllm.chat._think_kw': ('chat.html#_think_kw', 'fastllm/chat.py'),
-                              'fastllm.chat._trunc_content': ('chat.html#_trunc_content', 'fastllm/chat.py'),
                               'fastllm.chat._trunc_param': ('chat.html#_trunc_param', 'fastllm/chat.py'),
-                              'fastllm.chat._trunc_str': ('chat.html#_trunc_str', 'fastllm/chat.py'),
                               'fastllm.chat._url2content': ('chat.html#_url2content', 'fastllm/chat.py'),
                               'fastllm.chat._usrtools': ('chat.html#_usrtools', 'fastllm/chat.py'),
                               'fastllm.chat.add_warning': ('chat.html#add_warning', 'fastllm/chat.py'),
@@ -139,17 +129,13 @@ d = { 'settings': { 'branch': 'main',
                               'fastllm.chat.cite_footnote': ('chat.html#cite_footnote', 'fastllm/chat.py'),
                               'fastllm.chat.contents': ('chat.html#contents', 'fastllm/chat.py'),
                               'fastllm.chat.extract_fence_call': ('chat.html#extract_fence_call', 'fastllm/chat.py'),
-                              'fastllm.chat.fmt2hist': ('chat.html#fmt2hist', 'fastllm/chat.py'),
-                              'fastllm.chat.hist2fmt': ('chat.html#hist2fmt', 'fastllm/chat.py'),
                               'fastllm.chat.lite_mk_func': ('chat.html#lite_mk_func', 'fastllm/chat.py'),
                               'fastllm.chat.mk_msg': ('chat.html#mk_msg', 'fastllm/chat.py'),
                               'fastllm.chat.mk_msgs': ('chat.html#mk_msgs', 'fastllm/chat.py'),
-                              'fastllm.chat.mk_tr_details': ('chat.html#mk_tr_details', 'fastllm/chat.py'),
                               'fastllm.chat.postproc': ('chat.html#postproc', 'fastllm/chat.py'),
                               'fastllm.chat.remove_cache_ckpts': ('chat.html#remove_cache_ckpts', 'fastllm/chat.py'),
                               'fastllm.chat.run_fence_tool': ('chat.html#run_fence_tool', 'fastllm/chat.py'),
                               'fastllm.chat.search_count': ('chat.html#search_count', 'fastllm/chat.py'),
-                              'fastllm.chat.split_tools': ('chat.html#split_tools', 'fastllm/chat.py'),
                               'fastllm.chat.stop_reason': ('chat.html#stop_reason', 'fastllm/chat.py'),
                               'fastllm.chat.stop_sequences': ('chat.html#stop_sequences', 'fastllm/chat.py'),
                               'fastllm.chat.structured': ('chat.html#structured', 'fastllm/chat.py')},
@@ -272,18 +258,8 @@ d = { 'settings': { 'branch': 'main',
                                'fastllm.types.Completion': ('types.html#completion', 'fastllm/types.py'),
                                'fastllm.types.Completion._repr_markdown_': ('types.html#completion._repr_markdown_', 'fastllm/types.py'),
                                'fastllm.types.Completion.cost': ('types.html#completion.cost', 'fastllm/types.py'),
-                               'fastllm.types.Msg': ('types.html#msg', 'fastllm/types.py'),
-                               'fastllm.types.Msg._repr_markdown_': ('types.html#msg._repr_markdown_', 'fastllm/types.py'),
-                               'fastllm.types.Part': ('types.html#part', 'fastllm/types.py'),
-                               'fastllm.types.Part._repr_markdown_': ('types.html#part._repr_markdown_', 'fastllm/types.py'),
-                               'fastllm.types.ToolCall': ('types.html#toolcall', 'fastllm/types.py'),
-                               'fastllm.types.ToolCall._repr_markdown_': ('types.html#toolcall._repr_markdown_', 'fastllm/types.py'),
                                'fastllm.types.Usage': ('types.html#usage', 'fastllm/types.py'),
-                               'fastllm.types._fetch_url_partial': ('types.html#_fetch_url_partial', 'fastllm/types.py'),
-                               'fastllm.types._trunc_strs': ('types.html#_trunc_strs', 'fastllm/types.py'),
                                'fastllm.types.approx_pricing': ('types.html#approx_pricing', 'fastllm/types.py'),
-                               'fastllm.types.data_url': ('types.html#data_url', 'fastllm/types.py'),
-                               'fastllm.types.display_list': ('types.html#display_list', 'fastllm/types.py'),
                                'fastllm.types.fn_schema': ('types.html#fn_schema', 'fastllm/types.py'),
                                'fastllm.types.get_api_key': ('types.html#get_api_key', 'fastllm/types.py'),
                                'fastllm.types.get_model_info': ('types.html#get_model_info', 'fastllm/types.py'),
@@ -292,11 +268,7 @@ d = { 'settings': { 'branch': 'main',
                                'fastllm.types.infer_api_name': ('types.html#infer_api_name', 'fastllm/types.py'),
                                'fastllm.types.is_deepseek_peak_hour': ('types.html#is_deepseek_peak_hour', 'fastllm/types.py'),
                                'fastllm.types.mk_completion': ('types.html#mk_completion', 'fastllm/types.py'),
-                               'fastllm.types.mk_tool_res_msg': ('types.html#mk_tool_res_msg', 'fastllm/types.py'),
                                'fastllm.types.model_prices_meta': ('types.html#model_prices_meta', 'fastllm/types.py'),
-                               'fastllm.types.part_txt': ('types.html#part_txt', 'fastllm/types.py'),
                                'fastllm.types.payload_kwargs': ('types.html#payload_kwargs', 'fastllm/types.py'),
                                'fastllm.types.register_model_info': ('types.html#register_model_info', 'fastllm/types.py'),
-                               'fastllm.types.resize_b64': ('types.html#resize_b64', 'fastllm/types.py'),
-                               'fastllm.types.sys_text': ('types.html#sys_text', 'fastllm/types.py'),
-                               'fastllm.types.url_mime': ('types.html#url_mime', 'fastllm/types.py')}}}
+                               'fastllm.types.resize_b64': ('types.html#resize_b64', 'fastllm/types.py')}}}

{python_fastllm-0.0.30 → python_fastllm-0.0.32}/fastllm/acomplete.py

@@ -3,8 +3,8 @@
 # AUTOGENERATED! DO NOT EDIT! File to edit: ../nbs/06_acomplete.ipynb.
 
 # %% auto #0
-__all__ = ['specs_path', 'ant_spec', 'oai_spec', 'gem_spec', 'vendor_mapping', 'api2spec', 'defaults', 'mk_client',
-           'ContextWindowExceededError', 'acomplete']
+__all__ = ['specs_path', 'ant_spec', 'oai_spec', 'gem_spec', 'vendor_mapping', 'api2spec', 'defaults', 'split_vendor',
+           'mk_client', 'ContextWindowExceededError', 'acomplete']
 
 # %% ../nbs/06_acomplete.ipynb #f2f57253
 import asyncio,json,httpx
@@ -16,6 +16,7 @@ from fastspec.oapi import *
 from fastspec.errors import APIError
 
 from .types import *
+from aidialog.msg_parts import Part, PartType, Msg, mk_tool_res_msg
 from .streaming import *
 from .openai_responses import *
 from .openai_chat import *
@@ -51,11 +52,22 @@ vendor_mapping = {
 # %% ../nbs/06_acomplete.ipynb #77d27ea7
 api2spec = {'openai':oai_spec, 'openai_chat':oai_spec, 'anthropic':ant_spec, 'gemini':gem_spec}
 
+# %% ../nbs/06_acomplete.ipynb #e3ed40fb
+def split_vendor(model):
+    "Split a `'vendor/model'` path into `(vendor_name, model)`: the prefix must name a known vendor or a registered api, else `(None, model)`."
+    if model and '/' in model:
+        v, m = model.split('/', 1)
+        if v in vendor_mapping or v in api_registry.apis: return v, m
+    return None, model
+
 # %% ../nbs/06_acomplete.ipynb #79075d95
 @flexicache()
 def mk_client(model=None, vendor_name=None, api_name=None, api_key=None, base_url=None, xtra_hdrs=None,
     timeout=httpx.Timeout(connect=30, read=300, write=30, pool=10)):
     err_msg = f"please pass a valid one vendor: {', '.join(list(vendor_mapping))} or pass `api_name`,`base_url` and `api_key`"
+    if not vendor_name and not api_name and not (base_url and api_key):
+        v, m = split_vendor(model)
+        if v in vendor_mapping: vendor_name, model = v, m
     if vendor_name:
         override_base_url = base_url
         try:
@@ -145,6 +157,10 @@ async def acomplete(msgs, model, api_name=None, vendor_name=None, api_key=None,
                     base_url=None, xtra_body=None, xtra_hdrs=None, stream=False,
                     stop_callables=None, retries=2, retry_delay=0.5, **kwargs):
     "Unified completion across different APIs."
+    if not vendor_name and not api_name and not (base_url and api_key):
+        v, m = split_vendor(model)
+        if v in vendor_mapping: vendor_name, model = v, m
+        elif v: api_name, model = v, m  # a registered transport api (e.g. claude_code): not an HTTP vendor
     if api_name == 'claude_code':
         if not stream: raise NotImplementedError("claude_code backend supports stream=True only for now")
         api = api_registry.apis[api_name]

{python_fastllm-0.0.30 → python_fastllm-0.0.32}/fastllm/anthropic.py

@@ -15,6 +15,7 @@ from fastcore.meta import *
 from fastspec.errors import api_error_from_event
 
 from .types import *
+from aidialog.msg_parts import Part, PartType, Msg, ToolCall, data_url
 from .streaming import *
 from .streaming import mk_acollect_stream
 

{python_fastllm-0.0.30 → python_fastllm-0.0.32}/fastllm/chat.py

@@ -3,33 +3,28 @@
 # AUTOGENERATED! DO NOT EDIT! File to edit: ../nbs/07_chat.ipynb.
 
 # %% auto #0
-__all__ = ['tool_dtls_tag', 're_tools', 'token_dtls_tag', 're_token', 'think_start', 'think_end', 're_think', 'effort',
-           'MediaUrl', 'remove_cache_ckpts', 'contents', 'stop_reason', 'mk_msg', 'FenceToolStop', 'extract_fence_call',
-           'split_tools', 'fmt2hist', 'mk_msgs', 'cite_footnote', 'postproc', 'lite_mk_func', 'ToolResponse',
-           'structured', 'StopResponse', 'FullResponse', 'search_count', 'UsageStats', 'AsyncChat',
-           'astream_with_complete', 'ChatCallback', 'DeepseekMsgsCallback', 'DeepseekPrefillCallback', 'add_warning',
-           'StopReasonCallback', 'run_fence_tool', 'FenceToolCallback', 'ToolReminderCallback', 'stop_sequences',
-           'StopSequencesCallback', 'mk_tr_details', 'hist2fmt', 'StreamFormatter', 'AsyncStreamFormatter',
-           'adisplay_stream']
+__all__ = ['effort', 'remove_cache_ckpts', 'contents', 'stop_reason', 'mk_msg', 'FenceToolStop', 'extract_fence_call', 'mk_msgs',
+           'cite_footnote', 'postproc', 'lite_mk_func', 'ToolResponse', 'structured', 'search_count', 'UsageStats',
+           'AsyncChat', 'astream_with_complete', 'ChatCallback', 'DeepseekMsgsCallback', 'DeepseekPrefillCallback',
+           'add_warning', 'StopReasonCallback', 'run_fence_tool', 'FenceToolCallback', 'ToolReminderCallback',
+           'stop_sequences', 'StopSequencesCallback', 'StreamFormatter', 'AsyncStreamFormatter', 'adisplay_stream']
 
 # %% ../nbs/07_chat.ipynb #d5a3bc1f
-import base64, json
 from typing import Optional,Callable
-from html import escape
 from toolslm.funccall import mk_ns, call_func, call_func_async, get_schema
 from fastcore.utils import *
 from fastcore.meta import delegates
 from dataclasses import dataclass
+import base64
 
 from .types import *
+from fastcore.xtras import detect_mime
+from aidialog.msg_parts import (Msg, Part, PartType, ToolCall, mk_tool_res_msg, mk_tr_details,
+    tool_info, usage_info, think_start, think_end, StopResponse, display_list, MediaUrl, url_mime, fmt2hist,
+    _mk_result_fence, _split_fence_msgs, _fence_back, _trunc_str)
 from .acomplete import *
 
-# %% ../nbs/07_chat.ipynb #1b75c262
-class MediaUrl(BasicRepr):
-    "Direct URL media reference"
-    def __init__(self, url, mime=None): self.url, self.mime = url, ifnone(mime, url_mime(url))
-
-# %% ../nbs/07_chat.ipynb #eb557831
+# %% ../nbs/07_chat.ipynb #4a37bae5
 def _mime2part_type(mime):
     "Map MIME string to canonical PartType"
     if mime.startswith('image/'): return PartType.input_image
@@ -49,7 +44,7 @@ def _url2content(o):
     mime = o.mime or url_mime(o.url)
     return Part(type=_mime2part_type(mime), text=o.url, data=dict(mime=mime))
 
-# %% ../nbs/07_chat.ipynb #48c78e48
+# %% ../nbs/07_chat.ipynb #43a55cda
 def _add_cache_control(
     msg,          # LiteLLM formatted msg
     ttl=None # Cache TTL: '5m' (default) or '1h'
@@ -78,6 +73,8 @@ def _mk_content(o):
     elif isinstance(o, MediaUrl): return _url2content(o)
     return o
 
+
+# %% ../nbs/07_chat.ipynb #48c78e48
 def contents(c):
     "Get Msg object from Completion."
     if not c.message: return ''
@@ -87,7 +84,7 @@ def stop_reason(c):
     if not c.finish_reason: return 'unk'
     return c.finish_reason
 
-# %% ../nbs/07_chat.ipynb #8bdd997c
+# %% ../nbs/07_chat.ipynb #4a1c44ea
 def mk_msg(
     content,      # Content: str, bytes (image), list of mixed content, or dict w 'role' and 'content' fields
     role="user",  # Message role if content isn't already a dict/Message
@@ -105,20 +102,8 @@ def mk_msg(
     msg = Msg(role=role, content=parts)
     return _add_cache_control(msg, ttl=ttl) if cache else msg
 
-# %% ../nbs/07_chat.ipynb #db466e1c
-tool_dtls_tag = "<details class='tool-usage-details' markdown='1'>"
-re_tools = re.compile(fr"^({tool_dtls_tag}\n*(?:<summary>(?P<summary>.*?)</summary>\n*)?\n*```json\n+(.*?)\n+```\n+</details>)",
-    flags=re.DOTALL|re.MULTILINE)
-token_dtls_tag = "<details class='token-usage-details' markdown='1'>"
-re_token = re.compile(fr"^{re.escape(token_dtls_tag)}\n*<summary>.*?</summary>\n*\n*`.*?`\n*\n*</details>\n?",
-    flags=re.DOTALL|re.MULTILINE)
-think_start,think_end = '<!--think_start-->','<!--think_end-->'
-re_think = re.compile(rf'{re.escape(think_start)}.*?{re.escape(think_end)}\n?', re.DOTALL)
-
 # %% ../nbs/07_chat.ipynb #be998131
-_fence_back = '`````'
 _fence_re = re.compile(f'^{_fence_back}(py|bash)\n(.*?)\n{_fence_back}$', re.DOTALL | re.MULTILINE)
-_result_re = re.compile(f'\n{_fence_back}result\n(.*?)\n{_fence_back}\n', re.DOTALL)
 _lang2tool = dict(py='python', bash='bash')
 
 class FenceToolStop:
@@ -136,83 +121,7 @@ def extract_fence_call(text):
     m = ms[-1]
     if not text[m.end():].strip(): return m.group(1), m.group(2)
 
-# %% ../nbs/07_chat.ipynb #215183bf
-@patch(as_prop=True)
-def text(self:Msg): return ''.join(p.text or '' for p in self.content if p.type == PartType.text)
-
-# %% ../nbs/07_chat.ipynb #1de7e4d2
-def _mk_result_fence(output): return f"\n{_fence_back}result\n{output}\n{_fence_back}\n"
-
-def _split_msg_on_fences(msg):
-    "Split an assistant Msg on result fences, return list of Msgs"
-    if msg.role != 'assistant': return [msg]
-    if not _result_re.search(msg.text): return [msg]
-    res, asst_parts, tool_parts = [], [], []
-    for msg_part in msg.content:
-        if msg_part.type == PartType.thinking: asst_parts.append(msg_part)
-        elif msg_part.type == PartType.tool_use: tool_parts.append(msg_part)
-        elif parts := _result_re.split(msg_part.text or ''):
-            for i,p in enumerate(parts):
-                if not p: continue
-                if i % 2 == 0: res.append(Msg(role='assistant', content=asst_parts+[Part(type=PartType.text, text=p.strip())]))
-                else:          res.append(Msg(role='user', content=[Part(type=PartType.text, text=_mk_result_fence(p))]))
-    if tool_parts: res.append(Msg(role='assistant', content=tool_parts))
-    return res
-
-def _split_fence_msgs(msgs):
-    "Split all assistant msgs on result fences for wire protocol"
-    res = []
-    for m in msgs: res.extend(_split_msg_on_fences(m))
-    return res
-
-# %% ../nbs/07_chat.ipynb #45ada210
-def _extract_tool_parts(text:str):
-    "Extract (tool_use_part, tool_result_part) from <details> json block"
-    try: d = json.loads(text.strip())
-    except: return None
-    call = d['call']
-    # Skip server tool calls in deserialization (round trip issues with Gemini/Anthropic)
-    if d.get('server'): return None
-    tu = Part(type=PartType.tool_use, text=None, data=dict(id=d['id'], name=call['function'], arguments=call['arguments']))
-    tr = Part(type=PartType.tool_result, text=str(d['result']), data={'id': d['id'], 'name': call['function']})
-    return tu, tr
-
-def split_tools(s):
-    "Split formatted output into (text, summary, tooljson) chunks"
-    return [(txt,summ,tj) for txt,_,summ,tj in chunked(re_tools.split(s.strip()), 4, pad=True)]
-
-# %% ../nbs/07_chat.ipynb #44060a78
-def fmt2hist(outp:str)->list[Msg]:
-    "Transform a formatted output string into fastllm canonical Msgs"
-    if token_dtls_tag in outp: outp = re_token.sub('', outp)
-    if think_start in outp: outp = re_think.sub('', outp)
-    if tool_dtls_tag not in outp:
-        msg = Msg(role='assistant', content=[Part(type=PartType.text, text=outp.strip() or '.')])
-        return _split_msg_on_fences(msg)
-    hist, asst_parts, tool_parts = [], [], []
-    def flush():
-        if tool_parts:
-            hist.append(Msg(role='assistant', content=asst_parts.copy()))
-            hist.append(Msg(role='tool',      content=tool_parts.copy()))
-            asst_parts.clear()
-            tool_parts.clear()
-    for txt,_,tj in split_tools(outp):
-        if txt and txt.strip():
-            if tool_parts: flush()
-            asst_parts.append(Part(type=PartType.text, text=txt.strip() or '.'))
-        if tj and (tp := _extract_tool_parts(tj)):
-            asst_parts.append(tp[0])
-            tool_parts.append(tp[1])
-    flush()
-    if asst_parts: hist.append(Msg(role='assistant', content=asst_parts))
-    result = []
-    for msg in hist:
-        if msg.role == 'assistant': result.extend(_split_msg_on_fences(msg))
-        else: result.append(msg)
-    if result[-1].role == 'tool': result.append(Msg(role='assistant', content=[Part(type=PartType.text, text='.')])) 
-    return result
-
-# %% ../nbs/07_chat.ipynb #8de5ce8d
+# %% ../nbs/07_chat.ipynb #01bd81de
 def _apply_cache_idxs(msgs, cache_idxs=[-1], ttl=None):
     "Add cache control to `cache_idxs` after filtering tool-role msgs"
     ms = [j for j,m in enumerate(msgs) if m.role != 'tool']
@@ -221,7 +130,7 @@ def _apply_cache_idxs(msgs, cache_idxs=[-1], ttl=None):
         except IndexError: continue
         _add_cache_control(msgs[idx], ttl)
 
-# %% ../nbs/07_chat.ipynb #85882c9c
+# %% ../nbs/07_chat.ipynb #fcbb9f5d
 def mk_msgs(
     msgs,                   # List of messages (each: str, bytes, list, Msg, or Completion)
     cache=False,            # Enable Anthropic caching
@@ -231,7 +140,7 @@ def mk_msgs(
     "Create a list of fastllm canonical Msgs."
     if not msgs: return []
     if not isinstance(msgs, list): msgs = [msgs]
-    msgs = L(msgs).map(lambda m: fmt2hist(m) if isinstance(m,str) and (tool_dtls_tag in m or token_dtls_tag in m) else [m]).concat()
+    msgs = L(msgs).map(lambda m: fmt2hist(m) if isinstance(m,str) and (tool_info in m or usage_info in m) else [m]).concat()
     res, role = [], 'user'
     for m in msgs:
         res.append(msg := remove_cache_ckpts(mk_msg(m, role=role)))
@@ -311,30 +220,8 @@ effort['x'] = 'max'
 def _mk_prefill(pf): return dict(text=pf)
 
 # %% ../nbs/07_chat.ipynb #dc17f844
-class StopResponse(str): pass
-class FullResponse(str): pass
-
 def _has_stop(tres_parts): return any(isinstance(p.text, StopResponse) for p in tres_parts)
 
-# %% ../nbs/07_chat.ipynb #f58ce348
-def _trunc_str(s, mx=2000, skip=10, replace="TRUNCATED"):
-    "Truncate `s` to `mx` chars max, adding `replace` if truncated; `mx=None` disables truncation"
-    if mx is None or isinstance_str(s, ('FullResponse','Safe','PrettyString')): return s
-    if not isinstance(s, str): s = str(s)
-    s = type(s)(s.rstrip())
-    if len(s)>2 and s[0]=='𝍁' and s[-1]=='𝍁':
-        s = s[1:-1]
-        if replace: return s
-    if mx is None or len(s)<=mx: return s
-    s = s[skip:mx-skip]
-    ss = s.split(' ')
-    if len(ss[-1])>150: ss[-1] = ss[-1][:5]
-    s = ' '.join(ss)
-    if skip: s = f"…{s}"
-    s = f"{s}…"
-    if replace: s = f"<{replace}>{s}</{replace}>"
-    return s
-
 # %% ../nbs/07_chat.ipynb #ca9e447e
 _final_prompt = dict(role="user", content="You have used all your tool calls for this turn. Please summarize your findings. If you did not complete your goal, tell the user what further work is needed. You may use tools again on the next user message.")
 
@@ -379,8 +266,8 @@ class UsageStats:
 
     def fmt(self):
         if not self.total_tokens: return ''
-        summ = f"${self.cost:.4f}" if self.cost else f"{self.total_tokens:,} tokens"
-        return f"\n\n{token_dtls_tag}<summary>{summ}</summary>\n\n`{self!r}`\n\n</details>\n"
+        d = {k:v for k,v in self.__dict__.items() if v and not k.startswith('_')}
+        return "\n\n" + fenced(dumps(d, ensure_ascii=False), usage_info) + "\n"
 
 # %% ../nbs/07_chat.ipynb #cb3d7e77
 class AsyncChat:
@@ -406,12 +293,17 @@ class AsyncChat:
         default_cbs=True          # Whether to include default callbacks
     ):
         "LiteLLM chat client."
+        if not vendor_name and not api_name and not (base_url and api_key):
+            v, model = split_vendor(model)
+            if v in vendor_mapping: vendor_name = v
+            elif v: api_name = v  # a registered transport api (e.g. claude_code)
         self.model = model
         hist,tools = mk_msgs(hist,cache,cache_idxs,ttl),listify(tools)
         if ns is None and tools: ns = mk_ns(tools)
         elif ns is None: ns = globals()
         self.tool_schemas = [lite_mk_func(t) for t in tools] if tools else None
         self.use = UsageStats()
+        self.last_req_use = None  # usage of the latest request only; `use` accumulates across a turn's tool-call steps
         store_attr(but='cbs')
         self.cbs = L()
         if default_cbs: self.add_cbs(defaults.chat_callbacks)
@@ -435,6 +327,7 @@ class AsyncChat:
     def _track(self, res):
         u = UsageStats.from_response(res)
         u.cost *= (1 + self.markup)
+        self.last_req_use = u
         self.use += u
 
     def add_cb(self, cb):
@@ -721,44 +614,13 @@ def _trunc_param(v, mx=40):
 
 # %% ../nbs/07_chat.ipynb #80c0abdb
 def _tc_summary(tr):
-    "Format tool call as func(params) → result string"
+    "Format tool call as a `func(params)→result` code span"
     params = ', '.join(f"{k}={_trunc_param(v)}" for k,v in tr.data['arguments'].items())
     res = f"→{_trunc_param(tr.text)}" if tr.text else ''
-    return '<code>'+escape(f"{tr.data['name']}({params}){res}")+'</code>'
-
-# %% ../nbs/07_chat.ipynb #80f344cc
-def _trunc_content(content, mx):
-    "Truncate tool result content, respecting '_full' flag"
-    if isinstance(content, dict) and '_full' in content and len(content)==1: return content['_full']
-    return _trunc_str(content, mx=mx)
-
-# %% ../nbs/07_chat.ipynb #3602a033
-def mk_tr_details(tr, mx=2000):
-    "Create <details> block for tool call as JSON; `mx=None` disables truncation"
-    args = {k:_trunc_str(v, mx=None if mx is None else mx*5) if isinstance(v, str) else v for k,v in tr.data['arguments'].items()}
-    res = {'id':tr.data['id'], 'server':tr.data.get('server', False),
-           'call':{'function': tr.data['name'], 'arguments': args},
-           'result':_trunc_content(tr.text, mx=mx),}
-    summ = f"<summary>{_tc_summary(tr)}</summary>"
-    return f"\n\n{tool_dtls_tag}\n{summ}\n\n```json\n{dumps(res, indent=2, ensure_ascii=False)}\n```\n\n</details>\n\n"
-
-# %% ../nbs/07_chat.ipynb #ae9dfdb5
-def hist2fmt(msgs:list[Msg], mx=2000)->str:
-    "Render assistant/tool `msgs` as one formatted output string, the inverse of `fmt2hist`"
-    tus, out = {}, []
-    for m in msgs:
-        if m.role == 'assistant':
-            for p in m.content:
-                if p.type == PartType.text and p.text: out.append(p.text.strip())
-                elif p.type == PartType.tool_use: tus[p.data['id']] = p
-        elif m.role == 'tool':
-            for p in m.content:
-                if p.type != PartType.tool_result: continue
-                tu = tus.get(p.data['id'])
-                d = dict(p.data, arguments=tu.data.get('arguments', {}) if tu else {})
-                out.append(mk_tr_details(Part(type=PartType.tool_result, text=p.text, data=d), mx=mx).strip())
-        else: raise ValueError(f"hist2fmt renders assistant and tool messages only, got {m.role!r}")
-    return '\n\n'.join(o for o in out if o)
+    txt = f"{tr.data['name']}({params}){res}"
+    ticks = '`'*(max(map(len, re.findall('`+', txt)), default=0)+1)
+    pad = ' ' if '`' in txt else ''
+    return f"{ticks}{pad}{txt}{pad}{ticks}"
 
 # %% ../nbs/07_chat.ipynb #f0d984ec
 class StreamFormatter:

{python_fastllm-0.0.30 → python_fastllm-0.0.32}/fastllm/gemini.py

@@ -15,6 +15,7 @@ from fastcore.meta import *
 from fastspec.errors import api_error_from_event
 
 from .types import *
+from aidialog.msg_parts import Part, PartType, Msg, ToolCall, sys_text, part_txt, data_url, url_mime
 from .streaming import *
 from .streaming import mk_acollect_stream
 

{python_fastllm-0.0.30 → python_fastllm-0.0.32}/fastllm/openai_chat.py

@@ -14,6 +14,7 @@ from fastcore.meta import *
 from fastspec.errors import api_error_from_event
 
 from .types import *
+from aidialog.msg_parts import Part, PartType, Msg, ToolCall, sys_text, part_txt, data_url
 from .streaming import *
 from .streaming import mk_acollect_stream
 from .openai_responses import norm_usage

{python_fastllm-0.0.30 → python_fastllm-0.0.32}/fastllm/openai_responses.py

@@ -15,6 +15,7 @@ from fastcore.meta import *
 from fastspec.errors import api_error_from_event
 
 from .types import *
+from aidialog.msg_parts import Part, PartType, Msg, ToolCall, mk_tool_res_msg, sys_text, part_txt, data_url
 from .streaming import *
 from .streaming import mk_acollect_stream
 

{python_fastllm-0.0.30 → python_fastllm-0.0.32}/fastllm/streaming.py

@@ -12,6 +12,7 @@ from fastcore.utils import *
 from fastcore.meta import delegates
 from fastspec.errors import *
 from .types import *
+from aidialog.msg_parts import Part, PartType, Msg, ToolCall, _fence_back
 
 # %% ../nbs/01_streaming.ipynb #400d628a
 @dataclass

{python_fastllm-0.0.30 → python_fastllm-0.0.32}/fastllm/types.py

@@ -3,13 +3,12 @@
 # AUTOGENERATED! DO NOT EDIT! File to edit: ../nbs/00_types.ipynb.
 
 # %% auto #0
-__all__ = ['PartType', 'FinishReason', 'api_registry', 'model_prices_url', 'haik45', 'sonn45', 'sonn', 'sonn46', 'sonn5',
-           'opus46', 'opus', 'gpt54', 'gpt54m', 'gpt55', 'codex54', 'codex54m', 'codex55', 'codex53spark',
-           'model_info_registry', 'modern_llm', 'deepseek_v4_common', 'mimo_v25_common', 'codex_pricing', 'sol',
-           'terra', 'luna', 'gpt56s', 'Part', 'Msg', 'ToolCall', 'display_list', 'Usage', 'Completion', 'APIRegistry',
-           'mk_completion', 'mk_tool_res_msg', 'fn_schema', 'sys_text', 'part_txt', 'data_url', 'url_mime',
-           'payload_kwargs', 'get_api_key', 'resize_b64', 'model_prices_meta', 'infer_api_name', 'get_model_meta',
-           'register_model_info', 'get_model_info', 'get_model_pricing', 'approx_pricing', 'is_deepseek_peak_hour']
+__all__ = ['FinishReason', 'api_registry', 'model_prices_url', 'haik45', 'sonn45', 'sonn', 'sonn46', 'sonn5', 'opus46', 'opus',
+           'gpt54', 'gpt54m', 'gpt55', 'codex54', 'codex54m', 'codex55', 'codex53spark', 'model_info_registry',
+           'modern_llm', 'deepseek_v4_common', 'mimo_v25_common', 'codex_pricing', 'sol', 'terra', 'luna', 'gpt56s',
+           'Usage', 'Completion', 'APIRegistry', 'mk_completion', 'fn_schema', 'payload_kwargs', 'get_api_key',
+           'resize_b64', 'model_prices_meta', 'infer_api_name', 'get_model_meta', 'register_model_info',
+           'get_model_info', 'get_model_pricing', 'approx_pricing', 'is_deepseek_peak_hour']
 
 # %% ../nbs/00_types.ipynb #b4d047fd
 import httpx, base64, io
@@ -18,86 +17,8 @@ from datetime import datetime, timezone
 from fastcore.net import urljson
 from fastcore.utils import *
 from PIL import Image as PImg
+from aidialog.msg_parts import Part, PartType, Msg, ToolCall, data_url
 
-# %% ../nbs/00_types.ipynb #e568bade
-@dataclass
-class Part:
-    "A normalized content part."
-    type: str
-    text: str = None
-    data: dict = None
-
-# %% ../nbs/00_types.ipynb #aa3cfc23
-PartType = str_enum('PartType', 'text', 'thinking', 'refusal', 'tool_use', 'server_tool_result', 'tool_result',
-                    'input_image', 'input_audio', 'input_video', 'input_file')
-
-# %% ../nbs/00_types.ipynb #2eeff103
-def _trunc_strs(o, n=200):
-    "Truncate str or dict"
-    if not o: return o
-    if isinstance(o,str) and len(o)>n: return o[:100]+'...'
-    if isinstance(o,dict): return {k: (v[:100]+'...' if isinstance(v,str) and len(v)>n else v) for k,v in o.items()}
-    return o
-
-@patch
-def _repr_markdown_(self: Part):
-    body = _trunc_strs(self.text) if self.text else ''
-    data = _trunc_strs(self.data)
-    return f"""**Part** (`{self.type}`)
-
-{body}
-
-<details markdown='1'>
-
-- data: `{data}`
-
-</details>"""
-
-# %% ../nbs/00_types.ipynb #afeb9eef
-@dataclass
-class Msg:
-    "A normalized message."
-    role: str
-    content: List[Part]
-
-    def _repr_markdown_(self):
-        return f"""**Msg**
-
-- role: `{self.role}`
-
-<contents>
-
-{'\n\n'.join(p._repr_markdown_() for p in self.content)}
-
-</contents>"""
-
-# %% ../nbs/00_types.ipynb #f6d04920
-@dataclass
-class ToolCall:
-    "Normalized tool call."
-    id: str
-    name: str
-    arguments: dict = field(default_factory=dict)
-    server: bool = False
-    extra: dict = field(default_factory=dict)
-
-# %% ../nbs/00_types.ipynb #509cfc99
-@patch
-def _repr_markdown_(self: ToolCall):
-    extra = _trunc_strs(self.extra)
-    return f"""🔧 **{self.name}**(`{self.arguments}`)
-
-<details markdown='1'>
-
-- id: `{self.id}`
-- server: `{self.server}`
-- extra: `{extra}`
-
-</details>"""
-
-def display_list(l): 
-    from IPython.display import Markdown, display
-    display(Markdown('\n\n'.join(o._repr_markdown_() for o in l)))
 
 # %% ../nbs/00_types.ipynb #802ad832
 @dataclass(frozen=True)
@@ -177,15 +98,6 @@ def mk_completion(resp, model, api_name, vendor_name):
         vendor_name=vendor_name,
         raw=resp)
 
-# %% ../nbs/00_types.ipynb #d5322db5
-def mk_tool_res_msg(tool_calls:list[ToolCall], results:list[str|list]):
-    'A util to prepare parallel tool call with str or media list results'
-    parts = []
-    for tc,res in zip(tool_calls, results):
-        data = dict(id=tc.id, name=tc.name, arguments=tc.arguments, server=tc.server)
-        parts.append(Part(type=PartType.tool_result, text=res, data=data))
-    return Msg(role="tool", content=parts)
-
 # %% ../nbs/00_types.ipynb #8a8e468b
 def fn_schema(t):
     "Extract (name, description, parameters) from any tool format."
@@ -196,46 +108,6 @@ def fn_schema(t):
         return t.get('name',''), t.get('description',''), t.get('parameters', t.get('input_schema',{}))
     return None
 
-# %% ../nbs/00_types.ipynb #d1d48d91
-def sys_text(system):
-    "Extract text from system (str or Part)."
-    if system is None: return None
-    return system if isinstance(system, str) else system.text
-
-def part_txt(p): return p.text if isinstance(p,Part) else p
-
-# %% ../nbs/00_types.ipynb #f3deb055
-@flexicache(time_policy(24*3600))
-def _fetch_url_partial(url, nbytes=512): 
-    "Fetch remote media bytes, optionally only first `nbytes`."
-    try:
-        with httpx.stream('GET', url, headers={'Range': f'bytes=0-{nbytes-1}'}, follow_redirects=True) as r:
-            if r.status_code not in (200, 206): return
-            return r.read()
-    except (httpx.HTTPError, httpx.InvalidURL): return
-
-# %% ../nbs/00_types.ipynb #70a9a0c3
-_ext_mime = {
-    '.jpg':'image/jpeg', '.jpeg':'image/jpeg', '.png':'image/png', '.gif':'image/gif', '.webp':'image/webp',
-    '.pdf':'application/pdf',
-    '.mp3':'audio/mpeg', '.wav':'audio/wav', '.ogg':'audio/ogg', '.flac':'audio/flac', '.m4a':'audio/mp4',
-    '.mp4':'video/mp4', '.mov':'video/quicktime', '.webm':'video/webm',
-}
-
-def data_url(url):
-    "Parse data:mime;base64,data URL into (mime, b64_data), or None."
-    if not isinstance(url, str) or not url.startswith('data:') or ',' not in url: return None
-    header, body = url.split(',', 1)
-    if ';base64' not in header or not body: return None
-    return header[5:].split(';',1)[0].strip() or 'application/octet-stream', body
-
-def url_mime(url, default='application/octet-stream'):
-    "Guess mime from URL extension, and optional bytes fallback."
-    if "youtube.com" in url or "youtu.be" in url: return "video/mp4"
-    ext = '.' + url.rsplit('.', 1)[-1].split('?')[0].lower() if '.' in url.split('?')[0].split('/')[-1] else ''
-    if (mime:=_ext_mime.get(ext)) is None: return detect_mime(_fetch_url_partial(url))
-    return ifnone(mime, default)
-
 # %% ../nbs/00_types.ipynb #28c698fe
 def payload_kwargs(msgs, model, stream=False, system=None, max_tokens=None, temperature=None, tools=None, tool_choice=None, reasoning_effort=None, web_search_options=None, stop_callables=None): pass
 

{python_fastllm-0.0.30 → python_fastllm-0.0.32}/pyproject.toml

@@ -15,7 +15,7 @@ classifiers = [
     "Programming Language :: Python :: 3",
     "Programming Language :: Python :: 3 :: Only",
 ]
-dependencies = ['fastspec>=0.0.11', 'toolslm', 'pillow']
+dependencies = ['fastcore>=2.1.11', 'aidialog>=0.0.3', 'fastspec>=0.0.11', 'toolslm', 'pillow']
 
 [project.urls]
 Repository = "https://github.com/AnswerDotAI/fastllm"

{python_fastllm-0.0.30 → python_fastllm-0.0.32}/python_fastllm.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: python-fastllm
-Version: 0.0.30
+Version: 0.0.32
 Author-email: Kerem Turgutlu <keremturgutlu@gmail.com>
 License: Apache-2.0
 Project-URL: Repository, https://github.com/AnswerDotAI/fastllm
@@ -9,6 +9,8 @@ Classifier: Programming Language :: Python :: 3
 Classifier: Programming Language :: Python :: 3 :: Only
 Requires-Python: >=3.10
 Description-Content-Type: text/markdown
+Requires-Dist: fastcore>=2.1.11
+Requires-Dist: aidialog>=0.0.3
 Requires-Dist: fastspec>=0.0.11
 Requires-Dist: toolslm
 Requires-Dist: pillow

python_fastllm-0.0.32/python_fastllm.egg-info/requires.txt

@@ -0,0 +1,5 @@
+fastcore>=2.1.11
+aidialog>=0.0.3
+fastspec>=0.0.11
+toolslm
+pillow

python_fastllm-0.0.30/fastllm/__init__.py

@@ -1 +0,0 @@
-__version__ = "0.0.30"

python_fastllm-0.0.30/python_fastllm.egg-info/requires.txt

@@ -1,3 +0,0 @@
-fastspec>=0.0.11
-toolslm
-pillow