python-fastllm 0.0.7__tar.gz → 0.0.9__tar.gz

{python_fastllm-0.0.7 → python_fastllm-0.0.9}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: python-fastllm
-Version: 0.0.7
+Version: 0.0.9
 Author-email: Kerem Turgutlu <keremturgutlu@gmail.com>
 License: Apache-2.0
 Project-URL: Repository, https://github.com/AnswerDotAI/fastllm

python_fastllm-0.0.9/fastllm/__init__.py

@@ -0,0 +1 @@
+__version__ = "0.0.9"

{python_fastllm-0.0.7 → python_fastllm-0.0.9}/fastllm/_modidx.py

@@ -10,6 +10,7 @@ d = { 'settings': { 'branch': 'main',
                                    'fastllm.acomplete._classify_error': ('acomplete.html#_classify_error', 'fastllm/acomplete.py'),
                                    'fastllm.acomplete._classify_error_stream': ( 'acomplete.html#_classify_error_stream',
                                                                                  'fastllm/acomplete.py'),
+                                   'fastllm.acomplete._debug_print': ('acomplete.html#_debug_print', 'fastllm/acomplete.py'),
                                    'fastllm.acomplete._is_ctx_exceeded': ('acomplete.html#_is_ctx_exceeded', 'fastllm/acomplete.py'),
                                    'fastllm.acomplete.acomplete': ('acomplete.html#acomplete', 'fastllm/acomplete.py'),
                                    'fastllm.acomplete.mk_client': ('acomplete.html#mk_client', 'fastllm/acomplete.py')},
@@ -31,6 +32,7 @@ d = { 'settings': { 'branch': 'main',
                                    'fastllm.anthropic.denorm_tool_use': ('anthropic.html#denorm_tool_use', 'fastllm/anthropic.py'),
                                    'fastllm.anthropic.denorm_user': ('anthropic.html#denorm_user', 'fastllm/anthropic.py'),
                                    'fastllm.anthropic.denorm_web_search': ('anthropic.html#denorm_web_search', 'fastllm/anthropic.py'),
+                                   'fastllm.anthropic.finalize_usage': ('anthropic.html#finalize_usage', 'fastllm/anthropic.py'),
                                    'fastllm.anthropic.get_hdrs': ('anthropic.html#get_hdrs', 'fastllm/anthropic.py'),
                                    'fastllm.anthropic.mk_payload': ('anthropic.html#mk_payload', 'fastllm/anthropic.py'),
                                    'fastllm.anthropic.norm_finish': ('anthropic.html#norm_finish', 'fastllm/anthropic.py'),
@@ -145,6 +147,7 @@ d = { 'settings': { 'branch': 'main',
                               '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')},
+            'fastllm.codex': {},
             'fastllm.gemini': { 'fastllm.gemini._gem_filter_sch': ('gemini.html#_gem_filter_sch', 'fastllm/gemini.py'),
                                 'fastllm.gemini._gem_part_type': ('gemini.html#_gem_part_type', 'fastllm/gemini.py'),
                                 'fastllm.gemini.acollect_stream': ('gemini.html#acollect_stream', 'fastllm/gemini.py'),
@@ -268,17 +271,20 @@ d = { 'settings': { 'branch': 'main',
                                'fastllm.types.ToolCall._repr_markdown_': ('types.html#toolcall._repr_markdown_', 'fastllm/types.py'),
                                'fastllm.types.Usage': ('types.html#usage', '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'),
                                '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.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.sys_text': ('types.html#sys_text', 'fastllm/types.py'),
                                'fastllm.types.url_mime': ('types.html#url_mime', 'fastllm/types.py')}}}

{python_fastllm-0.0.7 → python_fastllm-0.0.9}/fastllm/acomplete.py

@@ -3,7 +3,7 @@
 # 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', 'mk_client',
+__all__ = ['specs_path', 'ant_spec', 'oai_spec', 'gem_spec', 'vendor_mapping', 'api2spec', 'defaults', 'mk_client',
            'ContextWindowExceededError', 'acomplete']
 
 # %% ../nbs/06_acomplete.ipynb #f2f57253
@@ -98,6 +98,21 @@ async def _classify_error_stream(gen):
         async for x in gen: yield x
     except APIError as e: raise _classify_error(e) from e
 
+# %% ../nbs/06_acomplete.ipynb #f626a4e1
+defaults = SimpleNamespace(debug_mode=False)
+
+def _debug_print(model, api_name, vendor_name, payload, func):
+    "Pretty-print acomplete inputs when defaults.debug_mode is set"
+    from pprint import pformat
+    p = dict(payload)
+    if defaults.debug_mode == 'brief' and 'tools' in p:
+        p['tools'] = '; '.join(o.get('name', o.get('type', o)) for o in p['tools'])
+    print('━'*60)
+    print(f"\033[1;36mfastllm debug\033[0m  model={model} vendor={vendor_name} api={api_name} base_url={func.base_url} path={func.path}")
+    print('─'*60)
+    print(f"\033[1;33mpayload:\033[0m\n{pformat(p, width=120, sort_dicts=False)}")
+    print('━'*60)
+
 # %% ../nbs/06_acomplete.ipynb #2379ec94
 @delegates(payload_kwargs)
 async def acomplete(msgs, model, api_name=None, vendor_name=None, api_key=None, base_url=None, xtra_body=None, xtra_hdrs=None,
@@ -114,6 +129,7 @@ async def acomplete(msgs, model, api_name=None, vendor_name=None, api_key=None,
         if vendor_name == 'deepseek' and 'v4' in model:   payload['messages'][-1]['prefix'] = True
         if vendor_name == 'moonshot' and 'kimi' in model: payload['messages'][-1]['partial'] = True
     func = attrgetter(api.op_path[stream])(cli)
+    if defaults.debug_mode: _debug_print(model, api_name, vendor_name, payload, func)
     try: resp = await func(**payload)
     except APIError as e: raise _classify_error(e) from e
     if stream: return _classify_error_stream(api.acollect_stream(resp, model=model, vendor_name=vendor_name, stop_callables=stop_callables))

{python_fastllm-0.0.7 → python_fastllm-0.0.9}/fastllm/anthropic.py

@@ -1,7 +1,7 @@
 # AUTOGENERATED! DO NOT EDIT! File to edit: ../nbs/04_anthropic.ipynb.
 
 # %% auto #0
-__all__ = ['ant_tc_types', 'api_ns', 'norm_tool_call', 'norm_tool_calls', 'norm_usage', 'norm_finish', 'norm_parts',
+__all__ = ['ant_tc_types', 'norm_tool_call', 'norm_tool_calls', 'norm_usage', 'finalize_usage', 'norm_finish', 'norm_parts',
            'norm_sse_event', 'delta_index_fn', 'acollect_stream', 'denorm_tool_use', 'denorm_assistant', 'denorm_tool',
            'denorm_msgs', 'denorm_tool_schs', 'denorm_tool_choice', 'denorm_reasoning', 'denorm_web_search',
            'denorm_system', 'denorm_user', 'denorm_image', 'denorm_file', 'denorm_tool_result', 'mk_payload',
@@ -42,7 +42,16 @@ def norm_usage(resp):
     pt = int(usg.get("input_tokens", 0) or 0) + cached + cache_creation
     ct = int(usg.get("output_tokens", 0) or 0)
     return Usage(prompt_tokens=pt, completion_tokens=ct, total_tokens=pt + ct,
-                 cached_tokens=cached, cache_creation_tokens=cache_creation, raw=usg)
+                 cached_tokens=cached, cache_creation_tokens=cache_creation, reasoning_tokens=0, raw=usg)
+
+def finalize_usage(usg, parts):
+    "Adjust usage using finalized Anthropic content parts."
+    if not usg: return usg
+    rc = '\n'.join(p.text or '' for p in parts if p.type == PartType.thinking)
+    ct = int(usg.raw.get('output_tokens', usg.completion_tokens) or 0)
+    rt = min(int(len(rc.split())*1.5), ct) if rc else 0
+    return Usage(prompt_tokens=usg.prompt_tokens, completion_tokens=ct-rt, total_tokens=usg.prompt_tokens+ct,
+        cached_tokens=usg.cached_tokens, cache_creation_tokens=usg.cache_creation_tokens, reasoning_tokens=rt, raw=usg.raw)
 
 # %% ../nbs/04_anthropic.ipynb #7a8b1f8f
 def norm_finish(resp, tcs=None):
@@ -197,7 +206,7 @@ def denorm_reasoning(v):
 def denorm_web_search(v):
     "Map canonical web_search_options to Anthropic hosted web_search tool."
     _max_uses = {"low": 1, "medium": 5, "high": 10}
-    t = {"type": "web_search_20260209", "name": "web_search"}
+    t = {"type": "web_search_20250305", "name": "web_search"}
     if (typ := (v or {}).get("type")): t["type"] = typ
     if (s := (v or {}).get("search_context_size")):
         t["max_uses"] = _max_uses.get(s, 5)
@@ -286,13 +295,6 @@ def cost(usage, m):
     return cost
 
 # %% ../nbs/04_anthropic.ipynb #f7c0b989
-api_ns = dict(norm_tool_calls=norm_tool_calls,
-                norm_parts=norm_parts,
-                norm_finish=norm_finish,
-                norm_usage=norm_usage,
-                acollect_stream=acollect_stream,
-                mk_payload=mk_payload,
-                cost=cost,
-                get_hdrs=get_hdrs,
-                op_path=('messages.messages_post','messages.messages_post'))
-api_registry.register('anthropic', **api_ns)
+api_registry.register('anthropic', norm_tool_calls=norm_tool_calls, norm_parts=norm_parts, norm_finish=norm_finish, norm_usage=norm_usage,
+    finalize_usage=finalize_usage, acollect_stream=acollect_stream, mk_payload=mk_payload, cost=cost, get_hdrs=get_hdrs,
+    op_path=('messages.messages_post','messages.messages_post'))

{python_fastllm-0.0.7 → python_fastllm-0.0.9}/fastllm/chat.py

@@ -93,7 +93,7 @@ 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)}<summary>.*?</summary>\n*\n*`.*?`\n*\n*</details>\n?",
+re_token = re.compile(fr"^{re.escape(token_dtls_tag)}\n*<summary>.*?</summary>\n*\n*`.*?`\n*\n*</details>\n?",
                       flags=re.DOTALL|re.MULTILINE)
 
 # %% ../nbs/07_chat.ipynb #be998131
@@ -209,7 +209,9 @@ 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 else [m]).concat()
+    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()
     res, role = [], 'user'
     for m in msgs:
         res.append(msg := remove_cache_ckpts(mk_msg(m, role=role)))
@@ -297,9 +299,11 @@ def _has_stop(tres_parts): return any(isinstance(p.text, StopResponse) for p in
 def _trunc_str(s, mx=2000, skip=10, replace="TRUNCATED"):
     "Truncate `s` to `mx` chars max, adding `replace` if truncated"
     if not isinstance(s, str): s = str(s)
-    if len(s)>2 and s[0]=='𝍁' and s[-1]=='𝍁': return s[1:-1]
+    s = s.rstrip()
+    if len(s)>2 and s[0]=='𝍁' and s[-1]=='𝍁':
+        s = s[1:-1]
+        if replace: return s
     if isinstance_str(s, ('FullResponse','Safe')): return s
-    s = str(s).strip()
     if len(s)<=mx: return s
     s = s[skip:mx-skip]
     ss = s.split(' ')
@@ -431,7 +435,8 @@ def _think_kw(model, think, vendor_name):
     if not think: return {}
     if 'opus-4-7' in model:
         e = 'xhigh' if think=='h' else effort.get(think)
-        return dict(thinking={"type":"adaptive", "display":"summarized"}, output_config={"effort":e})
+        eff = dict(thinking={"type":"adaptive", "display":"summarized"}, output_config={"effort":e})
+        return dict(reasoning_effort=eff)
     try: xhigh = get_model_info(model, vendor_name).get('supports_xhigh_reasoning_effort')
     except: xhigh = False
     eff = effort.get(think) if think!='x' else 'xhigh' if xhigh else 'high'
@@ -691,7 +696,7 @@ defaults.chat_callbacks = [DeepseekPrefillCallback, FenceToolCallback, ToolRemin
 def _trunc_param(v, mx=40):
     "Truncate and escape param value for display"
     tp = _trunc_str(str(v).replace('`', r'\`'), mx=mx, replace=None, skip=0)
-    try: return ast.literal_eval(tp)
+    try: return dumps(tp, ensure_ascii=False)
     except Exception: return repr(tp).replace('\\\\', '\\')
 
 # %% ../nbs/07_chat.ipynb #80c0abdb
@@ -721,7 +726,7 @@ def mk_tr_details(tr, mx=2000):
            '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)}\n```\n\n</details>\n\n"
+    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 #3049001c
 def mk_srv_tc_details(tc, mx=2000):
@@ -729,7 +734,7 @@ def mk_srv_tc_details(tc, mx=2000):
     args = {k:_trunc_str(v, mx=mx*5) for k,v in tc.arguments.items()}
     res = {'id':tc.id, 'server':True, 'call':{'function': tc.name, 'arguments': args}, 'result':"Server tool call executed."}
     summ = f"<summary>{_srv_tc_summary(tc)}</summary>"
-    return f"\n\n{tool_dtls_tag}\n{summ}\n\n```json\n{dumps(res, indent=2)}\n```\n\n</details>\n\n"
+    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 #f0d984ec
 # status_re = re.compile(r'^- ⏳ <code>(.*)</code> ⏳$|^🧠+$', re.MULTILINE) # TODO: Need to yield tool calls as they are done collated in fastllm `_acollect_stream`

python_fastllm-0.0.9/fastllm/codex.py

@@ -0,0 +1,7 @@
+# AUTOGENERATED! DO NOT EDIT! File to edit: ../nbs/05_codex.ipynb.
+
+# %% auto #0
+__all__ = []
+
+# %% ../nbs/05_codex.ipynb #a1d088d2
+from fastcore.utils import *

{python_fastllm-0.0.7 → python_fastllm-0.0.9}/fastllm/streaming.py

@@ -138,6 +138,7 @@ async def mk_acollect_stream(it, index_fn, model=None, api_name=None, vendor_nam
         deltas.append(d)
     part_accum.finalize()
     tcs = part_accum.tool_calls
+    if api_name: usg = api_registry.apis[api_name].finalize_usage(usg, part_accum.parts)
     if stop: fin = FinishReason.stop
     fin = FinishReason.tool_calls if fin==FinishReason.stop and any(~L(tcs).attrgot('server')) else fin # recheck tool calls post collation
     # tool calls and non-anthropic citations are yielded at the end
@@ -145,3 +146,4 @@ async def mk_acollect_stream(it, index_fn, model=None, api_name=None, vendor_nam
             message=Msg(role="assistant", content=part_accum.parts),
             finish_reason=fin, usage=usg, tool_calls=tcs, api_name=api_name, vendor_name=vendor_name,
             raw={'deltas':deltas})
+

{python_fastllm-0.0.7 → python_fastllm-0.0.9}/fastllm/types.py

@@ -4,10 +4,11 @@
 
 # %% auto #0
 __all__ = ['PartType', 'FinishReason', 'api_registry', 'model_prices_url', 'haik45', 'sonn45', 'sonn', 'sonn46', 'opus46', 'opus',
-           'gpt54', 'gpt54m', 'codex54', 'codex55', 'codex53spark', 'codex_pricing', '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', 'model_prices_meta',
-           'infer_api_name', 'get_model_meta', 'get_model_info']
+           'gpt54', 'gpt54m', 'gpt55', 'codex54', 'codex54m', 'codex55', 'codex53spark', 'model_info_registry',
+           'deepseek_v4_common', 'codex_pricing', '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', 'model_prices_meta', 'infer_api_name', 'get_model_meta',
+           'register_model_info', 'get_model_info', 'get_model_pricing', 'approx_pricing']
 
 # %% ../nbs/00_types.ipynb #b4d047fd
 from dataclasses import dataclass, field
@@ -147,25 +148,29 @@ FinishReason = str_enum('finish_reason', 'stop', 'tool_calls', 'length', 'conten
 # %% ../nbs/00_types.ipynb #fc681c52
 class APIRegistry:
     def __init__(self): self.apis = {}
-    def register(self, name, **kwargs): self.apis[name] = SimpleNamespace(**kwargs)
+    def register(self, name, finalize_usage=noop, **kwargs): self.apis[name] = SimpleNamespace(finalize_usage=finalize_usage, **kwargs)
 
 api_registry = APIRegistry()
 
+
 # %% ../nbs/00_types.ipynb #d58a5f96
 def mk_completion(resp, model, api_name, vendor_name):
     "Normalize an api response into Completion."
     api = api_registry.apis[api_name]
     tcs = api.norm_tool_calls(resp)
+    parts = api.norm_parts(resp)
+    usg = api.finalize_usage(api.norm_usage(resp), parts)
     return Completion(
         model=resp.get("model") or model,
-        message=Msg(role="assistant", content=api.norm_parts(resp)),
+        message=Msg(role="assistant", content=parts),
         finish_reason=api.norm_finish(resp, tcs),
-        usage=api.norm_usage(resp),
+        usage=usg,
         tool_calls=tcs,
         api_name=api_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'
@@ -225,6 +230,7 @@ def get_api_key(api_key, default):
 
 # %% ../nbs/00_types.ipynb #852adecd
 model_prices_url = 'https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json'
+
 @flexicache(time_policy(24*60*60))
 def model_prices_meta(): return urljson(model_prices_url)
 
@@ -254,63 +260,94 @@ opus46 = "claude-opus-4-6"
 opus = "claude-opus-4-7"
 gpt54 = "gpt-5.4"
 gpt54m = "gpt-5.4-mini"
+gpt55 = "gpt-5.5"
 codex54 = "gpt-5.4"
+codex54m = "gpt-5.4-mini"
 codex55 = "gpt-5.5"
 codex53spark = "gpt-5.3-codex-spark"
 
-# %% ../nbs/00_types.ipynb #d6d5b98c
-codex_pricing = {
-    "input_cost_per_token": 0.10 / 1_000_000,
-    "cache_creation_input_token_cost": 0.10 / 1_000_000,
-    "cache_read_input_token_cost": 0.10 / 1_000_000,
-    "output_cost_per_token": 0.50 / 1_000_000,
-}
+# %% ../nbs/00_types.ipynb #583e017b
+model_info_registry = {}
 
-_codex_overrides = {
-    codex53spark: dict(
-        supports_vision=False, supports_image_input=False, supports_web_search=True, supports_reasoning=True,
-        max_tokens=128000, max_input_tokens=128000, max_output_tokens=128000)
-}
+def register_model_info(model, vendor_name=None, base=None, base_vendor_name=None, **overrides):
+    "Register model metadata, optionally starting from `base`."
+    info = dict(get_model_info(base, base_vendor_name or vendor_name)) if base else {}
+    info.update(overrides)
+    model_info_registry[vendor_name, model] = info
 
-# %% ../nbs/00_types.ipynb #fbfdeb0a
-def get_model_info(mn, vendor_name=None, strict=False):
-    info = get_model_meta(mn, 'chatgpt' if vendor_name=='codex' else vendor_name)
-    # anthropic web search 
+def get_model_info(mn, vendor_name=None):
+    info = model_info_registry.get((vendor_name, mn)) or get_model_meta(mn, vendor_name)
     if 'search_context_cost_per_query' in info: info['supports_web_search'] = True
-    # kimi
-    if 'kimi' in mn: 
-        if 'k2p6' in mn: info = get_model_meta(mn.replace('k2p6', 'k2p5'), vendor_name)
-        info['supports_reasoning'] = True
-        info['supports_vision'] = True
-        if vendor_name == 'moonshot': info['supports_assistant_prefill'] = True
-    # gpt web search
-    if mn in ("gpt-5.4", "gpt-5.4-mini"):
-        info['supports_web_search'] = True
-        info.pop('mode', None)
-    # codex updates
-    if vendor_name == 'codex':
-        info = merge(info, codex_pricing)
-        info |= _codex_overrides.get(mn, {})
-    # deepseek v4
-    if vendor_name == 'deepseek' and mn in ("deepseek-v4-flash", "deepseek-v4-pro"):
-        info = dict(get_model_meta("deepseek/deepseek-v3.2"))
-        info |= dict(supports_assistant_prefill=True, supports_function_calling=True, supports_prompt_caching=True,
-            supports_reasoning=True, supports_tool_choice=True)
-        info.update(input_cost_per_token=1.4e-07, input_cost_per_token_cache_hit=2.8e-09, output_cost_per_token=2.8e-07,
-            max_input_tokens=1048576, max_output_tokens=393216, max_tokens=393216)
-        if 'pro' in mn: info = {**info, 'input_cost_per_token': 4.35e-07, 'input_cost_per_token_cache_hit': 3.625e-09, 'output_cost_per_token': 8.7e-07}
-    # qwen 3p6
-    if vendor_name == 'fireworks_ai' and mn == 'accounts/fireworks/models/qwen3p6-plus':
-        info = dict(supports_vision=True, supports_reasoning=True, supports_function_calling=True, supports_tool_choice=True,
-                    supports_system_messages=True, supports_response_schema=True, supports_parallel_function_calling=True,
-                    supports_prompt_caching=True, supports_native_streaming=True, supports_native_structured_output=True,
-                    max_tokens=1000000, max_input_tokens=1000000, max_output_tokens=65536,
-                    input_cost_per_token=0.5e-6, cache_read_input_token_cost=0.1e-6, output_cost_per_token=3.0e-6)
-    
-    # unresolved models
-    if not info and not strict: info = info | codex_pricing
     return dict2obj(info)
 
+# %% ../nbs/00_types.ipynb #8261dcd0
+register_model_info('accounts/fireworks/models/qwen3p6-plus', vendor_name='fireworks_ai',
+    supports_vision=True, supports_reasoning=True, supports_function_calling=True, supports_tool_choice=True,
+    supports_system_messages=True, supports_response_schema=True, supports_parallel_function_calling=True,
+    supports_prompt_caching=True, supports_native_streaming=True, supports_native_structured_output=True,
+    max_tokens=1000000, max_input_tokens=1000000, max_output_tokens=65536,
+    input_cost_per_token=0.5e-6, cache_read_input_token_cost=0.1e-6, output_cost_per_token=3.0e-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,
+    output_cost_per_reasoning_token=9e-6, cache_read_input_token_cost=1.5e-7)
+
+for model in ('gpt-5.4', 'gpt-5.4-mini'):
+    register_model_info(model, vendor_name='openai', base=model, supports_web_search=True, mode=None)
+
+for model in ('kimi-k2.5', 'kimi-k2.6'):
+    register_model_info(model, vendor_name='moonshot', base=f'moonshot/{model}', base_vendor_name=None,
+        supports_reasoning=True, supports_vision=True, supports_assistant_prefill=True)
+
+register_model_info('gemini-3.1-flash-lite', vendor_name='gemini', base='gemini-3.1-flash-lite-preview')
+register_model_info('models/gemini-3.1-flash-lite', vendor_name='gemini', base='gemini-3.1-flash-lite-preview')
+
+for model in ('accounts/fireworks/models/kimi-k2p5', 'accounts/fireworks/models/kimi-k2p6'):
+    register_model_info(model, vendor_name='fireworks_ai', base=model.replace('k2p6', 'k2p5'),
+        supports_reasoning=True, supports_vision=True,
+        input_cost_per_token=0.95e-6, cache_read_input_token_cost=0.16e-6, output_cost_per_token=4.0e-6)
+
+# %% ../nbs/00_types.ipynb #948d55d0
+deepseek_v4_common = dict(
+    supports_assistant_prefill=True, supports_function_calling=True, supports_prompt_caching=True,
+    supports_reasoning=True, supports_tool_choice=True,
+    max_input_tokens=1048576, max_output_tokens=393216, max_tokens=393216)
+
+register_model_info('deepseek-v4-flash', vendor_name='deepseek', base='deepseek/deepseek-v3.2', **deepseek_v4_common,
+    input_cost_per_token=1.4e-07, input_cost_per_token_cache_hit=2.8e-09,
+    output_cost_per_token=2.8e-07, cache_read_input_token_cost=1.4e-07/10)
+register_model_info('deepseek-v4-pro', vendor_name='deepseek', base='deepseek/deepseek-v3.2', **deepseek_v4_common,
+    input_cost_per_token=4.35e-07, input_cost_per_token_cache_hit=3.625e-09,
+    output_cost_per_token=8.7e-07, cache_read_input_token_cost=4.35e-07/10)
+
+# %% ../nbs/00_types.ipynb #2c23d11e
+codex_pricing = dict(
+    input_cost_per_token = 0.10/1_000_000, output_cost_per_token = 0.50/1_000_000,
+    cache_creation_input_token_cost = 0.10/1_000_000, cache_read_input_token_cost = 0.10/1_000_000)
+
+for model in (codex54, codex54m, codex55):
+    register_model_info(model, 'codex', base=model, base_vendor_name='chatgpt', supports_web_search=True, **codex_pricing)
+
+register_model_info(codex53spark, 'codex', **codex_pricing,
+    supports_vision=False, supports_image_input=False, supports_web_search=True, supports_reasoning=True,
+    max_tokens=128000, max_input_tokens=128000, max_output_tokens=128000)
+
+
+# %% ../nbs/00_types.ipynb #24cc47ec
+def get_model_pricing(mn, vendor_name, million=True):
+    return {k:round(v * (1e6 if million else 1), 6)
+        for k,v in get_model_info(mn, vendor_name).items()
+        if 'cost' in k and isinstance(v,float) and 'priority' not in k}
+
+# %% ../nbs/00_types.ipynb #79304cd9
+def approx_pricing(nm, vendor_name, out=10, cache=80, inp=10, markup=0):
+    "Approx cost per million tokens with given output/cache/input proportions"
+    p = get_model_pricing(nm, vendor_name)
+    ic = p.get('cache_creation_input_token_cost', p['input_cost_per_token'])
+    res = (p['output_cost_per_token']*out + p['cache_read_input_token_cost']*cache + ic*inp) / (out+cache+inp)
+    if nm=='claude-opus-4-7': res *= 1.5
+    return res*(1+markup)
+
 # %% ../nbs/00_types.ipynb #8bfca02d
 @patch(as_prop=True)
 def cost(self:Completion):

{python_fastllm-0.0.7 → python_fastllm-0.0.9}/python_fastllm.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: python-fastllm
-Version: 0.0.7
+Version: 0.0.9
 Author-email: Kerem Turgutlu <keremturgutlu@gmail.com>
 License: Apache-2.0
 Project-URL: Repository, https://github.com/AnswerDotAI/fastllm

{python_fastllm-0.0.7 → python_fastllm-0.0.9}/python_fastllm.egg-info/SOURCES.txt

@@ -5,6 +5,7 @@ fastllm/_modidx.py
 fastllm/acomplete.py
 fastllm/anthropic.py
 fastllm/chat.py
+fastllm/codex.py
 fastllm/gemini.py
 fastllm/openai_chat.py
 fastllm/openai_responses.py

python_fastllm-0.0.7/fastllm/__init__.py

@@ -1 +0,0 @@
-__version__ = "0.0.7"