python-fastllm 0.0.23__tar.gz → 0.0.25__tar.gz

{python_fastllm-0.0.23 → python_fastllm-0.0.25}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: python-fastllm
-Version: 0.0.23
+Version: 0.0.25
 Author-email: Kerem Turgutlu <keremturgutlu@gmail.com>
 License: Apache-2.0
 Project-URL: Repository, https://github.com/AnswerDotAI/fastllm

python_fastllm-0.0.25/fastllm/__init__.py

@@ -0,0 +1 @@
+__version__ = "0.0.25"

{python_fastllm-0.0.23 → python_fastllm-0.0.25}/fastllm/_modidx.py

@@ -287,6 +287,7 @@ d = { 'settings': { 'branch': 'main',
                                'fastllm.types.get_model_meta': ('types.html#get_model_meta', 'fastllm/types.py'),
                                'fastllm.types.get_model_pricing': ('types.html#get_model_pricing', 'fastllm/types.py'),
                                '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'),

{python_fastllm-0.0.23 → python_fastllm-0.0.25}/fastllm/acomplete.py

@@ -149,7 +149,7 @@ async def acomplete(msgs, model, api_name=None, vendor_name=None, api_key=None,
         api = api_registry.apis[api_name]
         payload = api.mk_payload(msgs, model, stream=stream, **kwargs)
         async def _mk_gen():
-            async for o in api.acollect_stream(payload, model=model, vendor_name=vendor_name, stop_callables=stop_callables): yield o
+            async for o in api.acollect_stream(payload, model=model, vendor_name='claude_code', stop_callables=stop_callables): yield o
         return _retry_stream(_mk_gen, retries, retry_delay)
     cli, api_name, vendor_name = mk_client(model, vendor_name, api_name, api_key, base_url, xtra_hdrs)
     api = api_registry.apis[api_name]

{python_fastllm-0.0.23 → python_fastllm-0.0.25}/fastllm/chat.py

@@ -3,13 +3,13 @@
 # AUTOGENERATED! DO NOT EDIT! File to edit: ../nbs/07_chat.ipynb.
 
 # %% auto #0
-__all__ = ['tool_dtls_tag', 're_tools', 'token_dtls_tag', 're_token', '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', 'StreamFormatter',
-           'AsyncStreamFormatter', 'adisplay_stream']
+__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', 'StreamFormatter', 'AsyncStreamFormatter', 'adisplay_stream']
 
 # %% ../nbs/07_chat.ipynb #d5a3bc1f
 import asyncio, base64, json, mimetypes, random, string, ast, warnings
@@ -111,6 +111,8 @@ re_tools = re.compile(fr"^({tool_dtls_tag}\n*(?:<summary>(?P<summary>.*?)</summa
 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 = '`````'
@@ -182,6 +184,7 @@ def split_tools(s):
 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)
@@ -345,13 +348,13 @@ def search_count(r):
 
 # %% ../nbs/07_chat.ipynb #61395e0d
 class UsageStats:
-    def __init__(self, prompt_tokens=0, completion_tokens=0, total_tokens=0, cached_tokens=0, cache_creation_tokens=0, reasoning_tokens=0, web_search_requests=0, cost=0.0): store_attr()
+    def __init__(self, model='', prompt_tokens=0, completion_tokens=0, total_tokens=0, cached_tokens=0, cache_creation_tokens=0, reasoning_tokens=0, web_search_requests=0, cost=0.0): store_attr()
 
     @classmethod
     def from_response(cls, r):
         u = r.usage
         return cls(
-            prompt_tokens=u.prompt_tokens or 0, completion_tokens=u.completion_tokens or 0, total_tokens=u.total_tokens or 0,
+            model=r.model or '', prompt_tokens=u.prompt_tokens or 0, completion_tokens=u.completion_tokens or 0, total_tokens=u.total_tokens or 0,
             cached_tokens=u.cached_tokens or 0, cache_creation_tokens=u.cache_creation_tokens or 0, reasoning_tokens=u.reasoning_tokens or 0,
             web_search_requests=search_count(r), cost=r.cost)
 
@@ -359,7 +362,7 @@ class UsageStats:
         if other is None: return self
         return UsageStats(**{k: getattr(self, k, 0) + getattr(other, k, 0)
             for k in ('prompt_tokens', 'completion_tokens', 'total_tokens', 'cached_tokens', 'cache_creation_tokens', 'reasoning_tokens', 'web_search_requests', 'cost')
-        })
+        }, model=other.model or self.model)
     def __radd__(self, other): return self if other is None or other == 0 else self.__add__(other)
 
     def __repr__(self):
@@ -369,6 +372,7 @@ class UsageStats:
         if self.reasoning_tokens: parts.append(f"reasoning={self.reasoning_tokens:,}")
         if getattr(self, 'web_search_requests', None): parts.append(f"searches={self.web_search_requests}")
         if self.cost: parts.append(f"${self.cost:.4f}")
+        if self.model: parts.append(self.model.split('/')[-1])
         return ' | '.join(parts)
 
     def fmt(self):
@@ -740,35 +744,42 @@ def mk_tr_details(tr, mx=2000):
 # %% ../nbs/07_chat.ipynb #f0d984ec
 class StreamFormatter:
     def __init__(self, mx=2000, debug=False, showthink=False):
-        self.outp,self.tcs = '',{}
+        self.outp,self.tcs,self._in_think = '',{},False
         store_attr()
     
     def format_item(self, o):
         "Format a single item from the response stream."
         res = ''
         if self.debug: print(o)
+        is_think = isinstance(o, dict) and o.get('thinking')
+        if not is_think and self._in_think: res += f'\n\n</details>\n{think_end}\n\n'; self._in_think = False
         if isinstance(o, dict):
             if thk:=o.get('thinking'):
-                if self.showthink: res += thk
-                res+= '🧠' if not self.outp or self.outp[-1]=='🧠' else '\n\n🧠'
+                if self.showthink:
+                    if not self._in_think:
+                        res += f'\n{think_start}\n<details><summary>Thinking</summary>\n\n'
+                        self._in_think = True
+                    res += thk
+                else: res+= '🧠' if not self.outp or self.outp[-1]=='🧠' else '\n\n🧠'
             elif self.outp and self.outp[-1] == '🧠': res+= '\n\n'
             if txt:=o.get('text'): res+=f"\n\n{txt}" if res and res[-1] == '🧠' else txt
-        if isinstance(o, Part) and o.type==PartType.tool_use:
-            res += f"\n- ⏳ {_tc_summary(o)} ⏳\n"
-        if isinstance(o, Part) and o.type == PartType.tool_result:
-            res += mk_tr_details(o,mx=self.mx)
+        if isinstance(o, Part):
+            if   o.type==PartType.tool_use: res += f"\n- ⏳ {_tc_summary(o)} ⏳\n"
+            elif o.type==PartType.tool_result: res += mk_tr_details(o,mx=self.mx)
         self.outp+=res
         return res
     
     def format_stream(self, rs):
         "Format the response stream for markdown display."
         for o in rs: yield self.format_item(o)
+        if self._in_think: yield f'\n\n</details>\n{think_end}\n\n'
 
 # %% ../nbs/07_chat.ipynb #0cdd4d7c
 class AsyncStreamFormatter(StreamFormatter):
     async def format_stream(self, rs):
         "Format the response stream for markdown display."
         async for o in rs: yield self.format_item(o)
+        if self._in_think: yield f'\n\n</details>\n{think_end}\n\n'
 
 # %% ../nbs/07_chat.ipynb #944bcd25
 @delegates(AsyncStreamFormatter)

{python_fastllm-0.0.23 → python_fastllm-0.0.25}/fastllm/types.py

@@ -9,11 +9,12 @@ __all__ = ['PartType', 'FinishReason', 'api_registry', 'model_prices_url', 'haik
            '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']
+           'get_model_pricing', 'approx_pricing', 'is_deepseek_peak_hour']
 
 # %% ../nbs/00_types.ipynb #b4d047fd
 import httpx, base64, io
 from dataclasses import dataclass, field
+from datetime import datetime, timezone
 from fastcore.net import urljson
 from fastcore.utils import *
 from PIL import Image as PImg
@@ -331,7 +332,11 @@ register_model_info('accounts/fireworks/models/qwen3p6-plus', vendor_name='firew
 
 register_model_info('accounts/fireworks/models/glm-5p2', vendor_name='fireworks_ai', **modern_llm, supports_vision=False,
     max_tokens=1000000, max_input_tokens=1000000, max_output_tokens=65536,
-    input_cost_per_token=1.4e-6, cache_read_input_token_cost=0.26e-6, output_cost_per_token=4.4e-6)
+    input_cost_per_token=1.4e-6, cache_read_input_token_cost=0.14e-6, output_cost_per_token=4.4e-6)
+
+register_model_info('accounts/fireworks/routers/glm-5p2-fast', vendor_name='fireworks_ai', **modern_llm, supports_vision=False,
+    max_tokens=1000000, max_input_tokens=1000000, max_output_tokens=65536,
+    input_cost_per_token=2.8e-6, cache_read_input_token_cost=0.28e-6, output_cost_per_token=8.8e-6)
 
 register_model_info('gemini-3.5-flash', vendor_name='gemini', base='gemini-3-flash-preview',
     input_cost_per_token=1.5e-6, output_cost_per_token=9e-6,
@@ -397,6 +402,10 @@ register_model_info(codex53spark, 'codex', **codex_pricing,
     supports_vision=False, supports_image_input=False, supports_web_search=True, supports_reasoning=True, supports_function_calling=True,
     max_tokens=128000, max_input_tokens=128000, max_output_tokens=128000)
 
+# %% ../nbs/00_types.ipynb #5d3e4720
+for model in (haik45, sonn45, sonn46, opus46, opus):
+    register_model_info(model, 'claude_code', base=model, base_vendor_name='anthropic', **codex_pricing)
+
 # %% ../nbs/00_types.ipynb #24cc47ec
 def get_model_pricing(mn, vendor_name, million=True):
     return {k:round(v * (1e6 if million else 1), 6)
@@ -412,10 +421,18 @@ def approx_pricing(nm, vendor_name, out=10, cache=80, inp=10, markup=0):
     if nm in ('claude-opus-4-7','claude-opus-4-8','claude-fable-5'): res *= 1.5
     return res*(1+markup)
 
+# %% ../nbs/00_types.ipynb #d2a310fb
+def is_deepseek_peak_hour(dt=None):
+    "Check whether current UTC time is in DeepSeek peak pricing hours."
+    dt = dt or datetime.now(timezone.utc)
+    h = dt.hour + dt.minute/60
+    return 1 <= h < 4 or 6 <= h < 10
+
 # %% ../nbs/00_types.ipynb #8bfca02d
 @patch(as_prop=True)
 def cost(self:Completion):
     meta = dict2obj(get_model_info(self.model, self.vendor_name))
     api = api_registry.apis[self.api_name]
     if not hasattr(api, 'cost'): raise NotImplementedError(f"API: {self.api_name} doesn't have a registered `cost` function in ns")
-    return api.cost(self.usage, meta)
+    res = api.cost(self.usage, meta)
+    return res*2 if self.vendor_name=='deepseek' and self.model.startswith('deepseek-v4') and is_deepseek_peak_hour() else res

{python_fastllm-0.0.23 → python_fastllm-0.0.25}/python_fastllm.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: python-fastllm
-Version: 0.0.23
+Version: 0.0.25
 Author-email: Kerem Turgutlu <keremturgutlu@gmail.com>
 License: Apache-2.0
 Project-URL: Repository, https://github.com/AnswerDotAI/fastllm

python_fastllm-0.0.23/fastllm/__init__.py

@@ -1 +0,0 @@
-__version__ = "0.0.23"