python-fastllm 0.0.10__tar.gz → 0.0.11__tar.gz

{python_fastllm-0.0.10 → python_fastllm-0.0.11}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: python-fastllm
-Version: 0.0.10
+Version: 0.0.11
 Author-email: Kerem Turgutlu <keremturgutlu@gmail.com>
 License: Apache-2.0
 Project-URL: Repository, https://github.com/AnswerDotAI/fastllm

python_fastllm-0.0.11/fastllm/__init__.py

@@ -0,0 +1 @@
+__version__ = "0.0.11"

{python_fastllm-0.0.10 → python_fastllm-0.0.11}/fastllm/acomplete.py

@@ -38,6 +38,7 @@ vendor_mapping = {
     "codex":        ('openai', 'https://chatgpt.com/backend-api/codex', 'CODEX_AUTH_TOKEN', _codex_json),
     "moonshot":     ('openai_chat', "https://api.moonshot.ai/v1", "MOONSHOT_API_KEY"),
     "deepseek":     ('openai_chat', "https://api.deepseek.com/v1", "DEEPSEEK_API_KEY"),
+    "mimo":         ('openai_chat', "https://api.xiaomimimo.com/v1", "MIMO_API_KEY"),
     "openrouter":   ('openai_chat', "https://openrouter.ai/api/v1", "OPENROUTER_API_KEY"),
     "together":     ('openai_chat', "https://api.together.xyz/v1", "TOGETHER_API_KEY"),
     "fireworks_ai": ('openai_chat', "https://api.fireworks.ai/inference/v1", "FIREWORKS_API_KEY"),
@@ -82,7 +83,7 @@ def _is_ctx_exceeded(code, msg):
     if str(code or "").lower() == "context_length_exceeded": return True
     return any(s in m for s in ("exceed context limit", "maximum context length", "maximum context limit",
     "longer than the model's context length", "input tokens exceed the configured limit",
-    "exceeds the maximum number of tokens allowed", "prompt is too long"))
+    "exceeds the maximum number of tokens allowed", "prompt is too long", "exceeds the context window"))
 
 def _classify_error(exc):
     "Upgrade generic `APIError` to a specific subclass if applicable."

{python_fastllm-0.0.10 → python_fastllm-0.0.11}/fastllm/anthropic.py

@@ -287,7 +287,7 @@ def cost(usage, m):
     in_tok = raw['input_tokens']
     cache_read = raw.get('cache_read_input_tokens', 0)
     cc = raw.get('cache_creation', {}) or {}
-    cache_5m  = cc.get('ephemeral_5m_input_tokens', 0)
+    cache_5m  = cc.get('ephemeral_5m_input_tokens', raw.get('cache_creation_input_tokens', 0))
     cache_1h  = cc.get('ephemeral_1h_input_tokens', 0)
     cost  = in_tok     * m.input_cost_per_token
     cost += raw['output_tokens'] * m.output_cost_per_token

{python_fastllm-0.0.10 → python_fastllm-0.0.11}/fastllm/chat.py

@@ -116,7 +116,7 @@ re_token = re.compile(fr"^{re.escape(token_dtls_tag)}\n*<summary>.*?</summary>\n
 _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')
+_lang2tool = dict(py='pyrun', bash='bash')
 
 class FenceToolStop:
     def __init__(self, langs): self.langs = langs
@@ -204,6 +204,7 @@ def fmt2hist(outp:str)->list[Msg]:
     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
@@ -635,7 +636,7 @@ def _active_fence_langs(tool_schemas):
 async def run_fence_tool(lang, code, ns):
     "Run the mapped tool for `lang` with the code, return result fence"
     tname = _lang2tool[lang]
-    arg = dict(code=code) if lang == 'py' else dict(command=code)
+    arg = dict(code=code) if lang == 'py' else dict(cmd=code)
     res = _mk_tool_result(await call_func_async(tname, arg, ns=ns, raise_on_err=False))
     return _mk_result_fence(_trunc_str(str(res)))
 

{python_fastllm-0.0.10 → python_fastllm-0.0.11}/fastllm/openai_chat.py

@@ -195,7 +195,7 @@ def get_hdrs(api_key=None):
 # %% ../nbs/03_oai_chat.ipynb #f89e2bf6
 def cost(usage, m):
     raw = usage.raw
-    pd, cd = raw.get('prompt_tokens_details', {}), raw.get('completion_tokens_details', {})
+    pd,cd = raw.get('prompt_tokens_details') or {},raw.get('completion_tokens_details') or {}
     cached = pd.get('cached_tokens', 0)
     in_audio, out_audio = pd.get('audio_tokens', 0), cd.get('audio_tokens', 0)
     in_txt  = raw['prompt_tokens']     - cached - in_audio

{python_fastllm-0.0.10 → python_fastllm-0.0.11}/fastllm/streaming.py

@@ -157,7 +157,7 @@ async def mk_acollect_stream(it, index_fn, model=None, api_name=None, vendor_nam
     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
-    yield Completion(d.raw.get('model', model),
+    yield Completion(model,
             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-0.0.10 → python_fastllm-0.0.11}/fastllm/types.py

@@ -5,10 +5,11 @@
 # %% auto #0
 __all__ = ['PartType', 'FinishReason', 'api_registry', 'model_prices_url', 'haik45', 'sonn45', 'sonn', 'sonn46', 'opus46', 'opus',
            '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']
+           'modern_llm', 'deepseek_v4_common', 'mimo_v25_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
 import httpx
@@ -164,7 +165,7 @@ def mk_completion(resp, model, api_name, vendor_name):
     parts = api.norm_parts(resp)
     usg = api.finalize_usage(api.norm_usage(resp), parts)
     return Completion(
-        model=resp.get("model") or model,
+        model=model,
         message=Msg(role="assistant", content=parts),
         finish_reason=api.norm_finish(resp, tcs),
         usage=usg,
@@ -173,7 +174,6 @@ 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'
@@ -271,7 +271,7 @@ haik45 = "claude-haiku-4-5"
 sonn45 = "claude-sonnet-4-5"
 sonn = sonn46 = "claude-sonnet-4-6"
 opus46 = "claude-opus-4-6"
-opus = "claude-opus-4-7"
+opus = "claude-opus-4-8"
 gpt54 = "gpt-5.4"
 gpt54m = "gpt-5.4-mini"
 gpt55 = "gpt-5.5"
@@ -287,6 +287,8 @@ def register_model_info(model, vendor_name=None, base=None, base_vendor_name=Non
     "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)
+    if isinstance(c := info.get('search_context_cost_per_query'), (int,float)):
+        info['search_context_cost_per_query'] = {f'search_context_size_{s}':c for s in ('low','medium','high')}
     model_info_registry[vendor_name, model] = info
 
 def get_model_info(mn, vendor_name=None):
@@ -294,11 +296,17 @@ def get_model_info(mn, vendor_name=None):
     if 'search_context_cost_per_query' in info: info['supports_web_search'] = True
     return dict2obj(info)
 
+# %% ../nbs/00_types.ipynb #331c5d0a
+register_model_info("claude-opus-4-8", vendor_name='anthropic', base="claude-opus-4-6")
+
+# %% ../nbs/00_types.ipynb #b36178d4
+modern_llm = dict(supports_function_calling=True, supports_tool_choice=True, supports_prompt_caching=True,
+    supports_parallel_function_calling=True, supports_native_streaming=True, supports_native_structured_output=True,
+    supports_reasoning=True, supports_response_schema=True, supports_system_messages=True)
+
 # %% ../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,
+register_model_info('accounts/fireworks/models/qwen3p6-plus', vendor_name='fireworks_ai', **modern_llm,
+    supports_vision=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)
 
@@ -322,9 +330,7 @@ for model in ('accounts/fireworks/models/kimi-k2p5', 'accounts/fireworks/models/
         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,
+deepseek_v4_common = dict(**modern_llm, supports_assistant_prefill=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,
@@ -334,19 +340,26 @@ register_model_info('deepseek-v4-pro', vendor_name='deepseek', base='deepseek/de
     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)
 
+mimo_v25_common = dict(**modern_llm, supports_web_search=True, max_input_tokens=1048576, max_output_tokens=131072, max_tokens=131072)
+
+register_model_info('mimo-v2.5-pro', vendor_name='mimo', **mimo_v25_common, base='deepseek/deepseek-v4-pro',
+    input_cost_per_token=0.435e-6, output_cost_per_token=0.87e-6, cache_read_input_token_cost=0.0036e-6, search_context_cost_per_query=0.005)
+register_model_info('mimo-v2.5', vendor_name='mimo', **mimo_v25_common, base='deepseek/deepseek-v4',
+    input_cost_per_token=0.14e-6,  output_cost_per_token=0.28e-6, cache_read_input_token_cost=0.0028e-6, search_context_cost_per_query=0.005,
+    supports_vision=True, supports_image_input=True)
+
 # %% ../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(model, 'codex', base=model, base_vendor_name='chatgpt', supports_web_search=True, max_input_tokens=256000, **codex_pricing)
 
 register_model_info(codex53spark, 'codex', **codex_pricing,
-    supports_vision=False, supports_image_input=False, supports_web_search=True, supports_reasoning=True,
+    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 #24cc47ec
 def get_model_pricing(mn, vendor_name, million=True):
     return {k:round(v * (1e6 if million else 1), 6)
@@ -359,7 +372,7 @@ def approx_pricing(nm, vendor_name, out=10, cache=80, inp=10, markup=0):
     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
+    if nm in ('claude-opus-4-7','claude-opus-4-8'): res *= 1.5
     return res*(1+markup)
 
 # %% ../nbs/00_types.ipynb #8bfca02d

{python_fastllm-0.0.10 → python_fastllm-0.0.11}/python_fastllm.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: python-fastllm
-Version: 0.0.10
+Version: 0.0.11
 Author-email: Kerem Turgutlu <keremturgutlu@gmail.com>
 License: Apache-2.0
 Project-URL: Repository, https://github.com/AnswerDotAI/fastllm

python_fastllm-0.0.10/fastllm/__init__.py

@@ -1 +0,0 @@
-__version__ = "0.0.10"