Skip to content

GitLab

Unverified Commit 9b08d975 authored by Chayenne's avatar Chayenne Committed by GitHub
Browse files

[docs] Refactor, remove compiled results and add gpt-oss (#9613) 


Co-authored-by: default avatarzhaochenyang20 <zhaochenyang20@gmail.com>
parent a0a77d93
Hide whitespace changes
Inline Side-by-side
Showing with 166 additions and 611 deletions
docs/advanced_features/function_calling.ipynb
View file @ 9b08d975
......@@ -51,7 +51,8 @@
"- mistral: Mistral (e.g. mistralai/Mistral-7B-Instruct-v0.3, mistralai/Mistral-Nemo-Instruct-2407, mistralai/\n",
"Mistral-Nemo-Instruct-2407, mistralai/Mistral-7B-v0.3).\n",
"- qwen25: Qwen 2.5 (e.g. Qwen/Qwen2.5-1.5B-Instruct, Qwen/Qwen2.5-7B-Instruct) and QwQ (i.e. Qwen/QwQ-32B). Especially, for QwQ, we can enable the reasoning parser together with tool call parser, details about reasoning parser can be found in [reasoning parser](https://docs.sglang.ai/backend/separate_reasoning.html).\n",
"- deepseekv3: DeepSeek-v3 (e.g., deepseek-ai/DeepSeek-V3-0324).\n"
"- deepseekv3: DeepSeek-v3 (e.g., deepseek-ai/DeepSeek-V3-0324).\n",
"- gpt-oss: GPT-OSS (e.g., openai/gpt-oss-120b, openai/gpt-oss-20b, lmsys/gpt-oss-120b-bf16, lmsys/gpt-oss-20b-bf16). Note: The gpt-oss tool parser filters out analysis channel events and only preserves normal text. This can cause the content to be empty when explanations are in the analysis channel. To work around this, complete the tool round by returning tool results as role=\"tool\" messages, which enables the model to generate the final content."
]
},
{
......@@ -354,142 +355,6 @@
"print(final_response.choices[0].message.content)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Tool Choice Mode\n",
"\n",
"SGLang supports OpenAI's `tool_choice` parameter to control when and which tools the model should call. This feature is implemented using EBNF (Extended Backus-Naur Form) grammar to ensure reliable tool calling behavior.\n",
"\n",
"### Supported Tool Choice Options\n",
"\n",
"- **`tool_choice=\"required\"`**: Forces the model to call at least one tool\n",
"- **`tool_choice={\"type\": \"function\", \"function\": {\"name\": \"specific_function\"}}`**: Forces the model to call a specific function\n",
"\n",
"### Backend Compatibility\n",
"\n",
"Tool choice is fully supported with the **Xgrammar backend**, which is the default grammar backend (`--grammar-backend xgrammar`). However, it may not be fully supported with other backends such as `outlines`.\n",
"\n",
"### Example: Required Tool Choice"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"from openai import OpenAI\n",
"from sglang.utils import wait_for_server, print_highlight, terminate_process\n",
"from sglang.test.doc_patch import launch_server_cmd\n",
"\n",
"# Start a new server session for tool choice examples\n",
"server_process_tool_choice, port_tool_choice = launch_server_cmd(\n",
" \"python3 -m sglang.launch_server --model-path Qwen/Qwen2.5-7B-Instruct --tool-call-parser qwen25 --host 0.0.0.0\"\n",
")\n",
"wait_for_server(f\"http://localhost:{port_tool_choice}\")\n",
"\n",
"# Initialize client for tool choice examples\n",
"client_tool_choice = OpenAI(\n",
" api_key=\"None\", base_url=f\"http://0.0.0.0:{port_tool_choice}/v1\"\n",
")\n",
"model_name_tool_choice = client_tool_choice.models.list().data[0].id\n",
"\n",
"# Example with tool_choice=\"required\" - forces the model to call a tool\n",
"messages_required = [\n",
" {\"role\": \"user\", \"content\": \"Hello, what is the capital of France?\"}\n",
"]\n",
"\n",
"# Define tools\n",
"tools = [\n",
" {\n",
" \"type\": \"function\",\n",
" \"function\": {\n",
" \"name\": \"get_current_weather\",\n",
" \"description\": \"Get the current weather in a given location\",\n",
" \"parameters\": {\n",
" \"type\": \"object\",\n",
" \"properties\": {\n",
" \"city\": {\n",
" \"type\": \"string\",\n",
" \"description\": \"The city to find the weather for, e.g. 'San Francisco'\",\n",
" },\n",
" \"unit\": {\n",
" \"type\": \"string\",\n",
" \"description\": \"The unit to fetch the temperature in\",\n",
" \"enum\": [\"celsius\", \"fahrenheit\"],\n",
" },\n",
" },\n",
" \"required\": [\"city\", \"unit\"],\n",
" },\n",
" },\n",
" }\n",
"]\n",
"\n",
"response_required = client_tool_choice.chat.completions.create(\n",
" model=model_name_tool_choice,\n",
" messages=messages_required,\n",
" temperature=0,\n",
" max_tokens=1024,\n",
" tools=tools,\n",
" tool_choice=\"required\", # Force the model to call a tool\n",
")\n",
"\n",
"print_highlight(\"Response with tool_choice='required':\")\n",
"print(\"Content:\", response_required.choices[0].message.content)\n",
"print(\"Tool calls:\", response_required.choices[0].message.tool_calls)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Example: Specific Function Choice\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# Example with specific function choice - forces the model to call a specific function\n",
"messages_specific = [\n",
" {\"role\": \"user\", \"content\": \"What are the most attactive places in France?\"}\n",
"]\n",
"\n",
"response_specific = client_tool_choice.chat.completions.create(\n",
" model=model_name_tool_choice,\n",
" messages=messages_specific,\n",
" temperature=0,\n",
" max_tokens=1024,\n",
" tools=tools,\n",
" tool_choice={\n",
" \"type\": \"function\",\n",
" \"function\": {\"name\": \"get_current_weather\"},\n",
" }, # Force the model to call the specific get_current_weather function\n",
")\n",
"\n",
"print_highlight(\"Response with specific function choice:\")\n",
"print(\"Content:\", response_specific.choices[0].message.content)\n",
"print(\"Tool calls:\", response_specific.choices[0].message.tool_calls)\n",
"\n",
"if response_specific.choices[0].message.tool_calls:\n",
" tool_call = response_specific.choices[0].message.tool_calls[0]\n",
" print(f\"Called function: {tool_call.function.name}\")\n",
" print(f\"Arguments: {tool_call.function.arguments}\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"terminate_process(server_process_tool_choice)"
]
},
{
"cell_type": "markdown",
"metadata": {},
......@@ -583,6 +448,9 @@
" messages, tokenize=True, add_generation_prompt=True, tools=tools\n",
")\n",
"\n",
"# Note that for gpt-oss tool parser, adding \"no_stop_trim\": True\n",
"# to make sure the tool call token <call> is not trimmed.\n",
"\n",
"sampling_params = {\n",
" \"max_new_tokens\": 1024,\n",
" \"temperature\": 0,\n",
......@@ -636,6 +504,142 @@
"llm.shutdown()"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Tool Choice Mode\n",
"\n",
"SGLang supports OpenAI's `tool_choice` parameter to control when and which tools the model should call. This feature is implemented using EBNF (Extended Backus-Naur Form) grammar to ensure reliable tool calling behavior.\n",
"\n",
"### Supported Tool Choice Options\n",
"\n",
"- **`tool_choice=\"required\"`**: Forces the model to call at least one tool\n",
"- **`tool_choice={\"type\": \"function\", \"function\": {\"name\": \"specific_function\"}}`**: Forces the model to call a specific function\n",
"\n",
"### Backend Compatibility\n",
"\n",
"Tool choice is fully supported with the **Xgrammar backend**, which is the default grammar backend (`--grammar-backend xgrammar`). However, it may not be fully supported with other backends such as `outlines`.\n",
"\n",
"### Example: Required Tool Choice"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"from openai import OpenAI\n",
"from sglang.utils import wait_for_server, print_highlight, terminate_process\n",
"from sglang.test.doc_patch import launch_server_cmd\n",
"\n",
"# Start a new server session for tool choice examples\n",
"server_process_tool_choice, port_tool_choice = launch_server_cmd(\n",
" \"python3 -m sglang.launch_server --model-path Qwen/Qwen2.5-7B-Instruct --tool-call-parser qwen25 --host 0.0.0.0\"\n",
")\n",
"wait_for_server(f\"http://localhost:{port_tool_choice}\")\n",
"\n",
"# Initialize client for tool choice examples\n",
"client_tool_choice = OpenAI(\n",
" api_key=\"None\", base_url=f\"http://0.0.0.0:{port_tool_choice}/v1\"\n",
")\n",
"model_name_tool_choice = client_tool_choice.models.list().data[0].id\n",
"\n",
"# Example with tool_choice=\"required\" - forces the model to call a tool\n",
"messages_required = [\n",
" {\"role\": \"user\", \"content\": \"Hello, what is the capital of France?\"}\n",
"]\n",
"\n",
"# Define tools\n",
"tools = [\n",
" {\n",
" \"type\": \"function\",\n",
" \"function\": {\n",
" \"name\": \"get_current_weather\",\n",
" \"description\": \"Get the current weather in a given location\",\n",
" \"parameters\": {\n",
" \"type\": \"object\",\n",
" \"properties\": {\n",
" \"city\": {\n",
" \"type\": \"string\",\n",
" \"description\": \"The city to find the weather for, e.g. 'San Francisco'\",\n",
" },\n",
" \"unit\": {\n",
" \"type\": \"string\",\n",
" \"description\": \"The unit to fetch the temperature in\",\n",
" \"enum\": [\"celsius\", \"fahrenheit\"],\n",
" },\n",
" },\n",
" \"required\": [\"city\", \"unit\"],\n",
" },\n",
" },\n",
" }\n",
"]\n",
"\n",
"response_required = client_tool_choice.chat.completions.create(\n",
" model=model_name_tool_choice,\n",
" messages=messages_required,\n",
" temperature=0,\n",
" max_tokens=1024,\n",
" tools=tools,\n",
" tool_choice=\"required\", # Force the model to call a tool\n",
")\n",
"\n",
"print_highlight(\"Response with tool_choice='required':\")\n",
"print(\"Content:\", response_required.choices[0].message.content)\n",
"print(\"Tool calls:\", response_required.choices[0].message.tool_calls)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Example: Specific Function Choice\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# Example with specific function choice - forces the model to call a specific function\n",
"messages_specific = [\n",
" {\"role\": \"user\", \"content\": \"What are the most attactive places in France?\"}\n",
"]\n",
"\n",
"response_specific = client_tool_choice.chat.completions.create(\n",
" model=model_name_tool_choice,\n",
" messages=messages_specific,\n",
" temperature=0,\n",
" max_tokens=1024,\n",
" tools=tools,\n",
" tool_choice={\n",
" \"type\": \"function\",\n",
" \"function\": {\"name\": \"get_current_weather\"},\n",
" }, # Force the model to call the specific get_current_weather function\n",
")\n",
"\n",
"print_highlight(\"Response with specific function choice:\")\n",
"print(\"Content:\", response_specific.choices[0].message.content)\n",
"print(\"Tool calls:\", response_specific.choices[0].message.tool_calls)\n",
"\n",
"if response_specific.choices[0].message.tool_calls:\n",
" tool_call = response_specific.choices[0].message.tool_calls[0]\n",
" print(f\"Called function: {tool_call.function.name}\")\n",
" print(f\"Arguments: {tool_call.function.arguments}\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"terminate_process(server_process_tool_choice)"
]
},
{
"cell_type": "markdown",
"metadata": {},
......@@ -657,6 +661,8 @@
"\n",
"For more information, refer to Meta’s documentation on [Zero shot function calling](https://github.com/meta-llama/llama-models/blob/main/models/llama4/prompt_format.md#zero-shot-function-calling---system-message).\n",
"\n",
"Note that this feature is still under development on Blackwell.\n",
"\n",
"### How to enable\n",
"- Launch the server with `--tool-call-parser pythonic`\n",
"- You may also specify --chat-template with the improved template for the model (e.g., `--chat-template=examples/chat_template/tool_chat_template_llama4_pythonic.jinja`).\n",
......
docs/advanced_features/separate_reasoning.ipynb
View file @ 9b08d975
......@@ -17,7 +17,7 @@
"| [Standard Qwen3 models](https://huggingface.co/collections/Qwen/qwen3-67dd247413f0e2e4f653967f) | `<think>` … `</think>` | `qwen3` | Supports `enable_thinking` parameter |\n",
"| [Qwen3-Thinking models](https://huggingface.co/Qwen/Qwen3-235B-A22B-Thinking-2507) | `<think>` … `</think>` | `qwen3` or `qwen3-thinking` | Always generates thinking content |\n",
"| [Kimi models](https://huggingface.co/moonshotai/models) | `◁think▷` … `◁/think▷` | `kimi` | Uses special thinking delimiters |\n",
"\n",
"| [GPT OSS](https://huggingface.co/openai/gpt-oss-120b) | `<\\|channel\\|>analysis<\\|message\\|>` … `<\\|end\\|>` | `gpt-oss` | N/A |\n",
"### Model-Specific Behaviors\n",
"\n",
"**DeepSeek-R1 Family:**\n",
......
docs/advanced_features/vlm_query.ipynb
View file @ 9b08d975
This source diff could not be displayed because it is too large. You can view the blob instead.
docs/basic_usage/gpt_oss.md
View file @ 9b08d975
......@@ -23,6 +23,11 @@ GPT‑OSS can call built‑in tools for web search and Python execution. You can
- Uses the Exa backend for web search.
- Requires an Exa API key; set `EXA_API_KEY` in your environment. Create a key at `https://exa.ai`.
### Tool & Reasoning Parser
- We support OpenAI Reasoning and Tool Call parser, as well as our SGLang native api for tool call and reasoning. Refer to [reasoning parser](../advanced_features/separate_reasoning.ipynb) and [tool call parser](../advanced_features/function_calling.ipynb) for more details.
## Notes
- Use **Python 3.12** for the demo tools. And install the required `gpt-oss` packages.
......
scripts/playground/frontend_reasoning.ipynb
View file @ 9b08d975
......@@ -13,63 +13,7 @@
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"/workspaces/sglang/.venv/lib/python3.12/site-packages/tqdm/auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n",
" from .autonotebook import tqdm as notebook_tqdm\n"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"[2025-05-05 17:53:32] server_args=ServerArgs(model_path='Qwen/Qwen3-4B', tokenizer_path='Qwen/Qwen3-4B', tokenizer_mode='auto', skip_tokenizer_init=False, enable_tokenizer_batch_encode=False, load_format='auto', trust_remote_code=False, dtype='auto', kv_cache_dtype='auto', quantization=None, quantization_param_path=None, context_length=None, device='cuda', served_model_name='Qwen/Qwen3-4B', chat_template=None, completion_template=None, is_embedding=False, revision=None, host='0.0.0.0', port=38475, mem_fraction_static=0.88, max_running_requests=None, max_total_tokens=None, chunked_prefill_size=8192, max_prefill_tokens=16384, schedule_policy='fcfs', schedule_conservativeness=1.0, cpu_offload_gb=0, page_size=1, tp_size=1, pp_size=1, max_micro_batch_size=None, stream_interval=1, stream_output=False, random_seed=376691526, constrained_json_whitespace_pattern=None, watchdog_timeout=300, dist_timeout=None, download_dir=None, base_gpu_id=0, gpu_id_step=1, log_level='info', log_level_http=None, log_requests=False, log_requests_level=0, show_time_cost=False, enable_metrics=False, decode_log_interval=40, api_key=None, file_storage_path='sglang_storage', enable_cache_report=False, reasoning_parser='qwen3', dp_size=1, load_balance_method='round_robin', ep_size=1, dist_init_addr=None, nnodes=1, node_rank=0, json_model_override_args='{}', lora_paths=None, max_loras_per_batch=8, lora_backend='triton', attention_backend=None, sampling_backend='flashinfer', grammar_backend='xgrammar', speculative_algorithm=None, speculative_draft_model_path=None, speculative_num_steps=None, speculative_eagle_topk=None, speculative_num_draft_tokens=None, speculative_accept_threshold_single=1.0, speculative_accept_threshold_acc=1.0, speculative_token_map=None, enable_double_sparsity=False, ds_channel_config_path=None, ds_heavy_channel_num=32, ds_heavy_token_num=256, ds_heavy_channel_type='qk', ds_sparse_decode_threshold=4096, disable_radix_cache=False, disable_cuda_graph=False, disable_cuda_graph_padding=False, enable_nccl_nvls=False, disable_outlines_disk_cache=False, disable_custom_all_reduce=False, enable_multimodal=None, disable_overlap_schedule=False, enable_mixed_chunk=False, enable_dp_attention=False, enable_ep_moe=False, enable_deepep_moe=False, deepep_mode='auto', enable_torch_compile=False, torch_compile_max_bs=32, cuda_graph_max_bs=None, cuda_graph_bs=None, torchao_config='', enable_nan_detection=False, enable_p2p_check=False, triton_attention_reduce_in_fp32=False, triton_attention_num_kv_splits=8, num_continuous_decode_steps=1, delete_ckpt_after_loading=False, enable_memory_saver=False, allow_auto_truncate=False, enable_custom_logit_processor=False, tool_call_parser=None, enable_hierarchical_cache=False, hicache_ratio=2.0, hicache_size=0, hicache_write_policy='write_through_selective', flashinfer_mla_disable_ragged=False, warmups=None, moe_dense_tp_size=None, n_share_experts_fusion=0, disable_chunked_prefix_cache=False, disable_fast_image_processor=False, debug_tensor_dump_output_folder=None, debug_tensor_dump_input_file=None, debug_tensor_dump_inject=False, disaggregation_mode='null', disaggregation_bootstrap_port=8998, disaggregation_transfer_backend='mooncake', disaggregation_ib_device=None)\n",
"[2025-05-05 17:53:38] Attention backend not set. Use flashinfer backend by default.\n",
"[2025-05-05 17:53:38] Init torch distributed begin.\n",
"[2025-05-05 17:53:38] Init torch distributed ends. mem usage=0.00 GB\n",
"[2025-05-05 17:53:38] Load weight begin. avail mem=43.89 GB\n",
"[2025-05-05 17:53:39] Using model weights format ['*.safetensors']\n",
"Loading safetensors checkpoint shards: 0% Completed | 0/3 [00:00<?, ?it/s]\n",
"Loading safetensors checkpoint shards: 67% Completed | 2/3 [00:00<00:00, 4.06it/s]\n",
"Loading safetensors checkpoint shards: 100% Completed | 3/3 [00:01<00:00, 2.52it/s]\n",
"Loading safetensors checkpoint shards: 100% Completed | 3/3 [00:01<00:00, 2.73it/s]\n",
"\n",
"[2025-05-05 17:53:40] Load weight end. type=Qwen3ForCausalLM, dtype=torch.bfloat16, avail mem=36.25 GB, mem usage=7.63 GB.\n",
"[2025-05-05 17:53:40] KV Cache is allocated. #tokens: 225647, K size: 15.49 GB, V size: 15.49 GB\n",
"[2025-05-05 17:53:40] Memory pool end. avail mem=4.71 GB\n",
"2025-05-05 17:53:41,152 - INFO - flashinfer.jit: Prebuilt kernels not found, using JIT backend\n",
"[2025-05-05 17:53:41] Capture cuda graph begin. This can take up to several minutes. avail mem=4.09 GB\n",
"[2025-05-05 17:53:41] Capture cuda graph bs [1, 2, 4, 8, 16, 24, 32, 40, 48, 56, 64, 72, 80, 88, 96, 104, 112, 120, 128, 136, 144, 152, 160]\n",
"Capturing batches (avail_mem=4.06 GB): 0%| | 0/23 [00:00<?, ?it/s]2025-05-05 17:53:41,620 - INFO - flashinfer.jit: Loading JIT ops: batch_decode_with_kv_cache_dtype_q_bf16_dtype_kv_bf16_dtype_o_bf16_dtype_idx_i32_head_dim_qk_128_head_dim_vo_128_posenc_0_use_swa_False_use_logits_cap_False\n",
"2025-05-05 17:53:41,642 - INFO - flashinfer.jit: Finished loading JIT ops: batch_decode_with_kv_cache_dtype_q_bf16_dtype_kv_bf16_dtype_o_bf16_dtype_idx_i32_head_dim_qk_128_head_dim_vo_128_posenc_0_use_swa_False_use_logits_cap_False\n",
"Capturing batches (avail_mem=2.68 GB): 100%|██████████| 23/23 [00:06<00:00, 3.75it/s]\n",
"[2025-05-05 17:53:47] Capture cuda graph end. Time elapsed: 6.18 s. mem usage=1.41 GB. avail mem=2.67 GB.\n",
"[2025-05-05 17:53:47] max_total_num_tokens=225647, chunked_prefill_size=8192, max_prefill_tokens=16384, max_running_requests=2821, context_len=40960\n",
"[2025-05-05 17:53:48] INFO: Started server process [1104179]\n",
"[2025-05-05 17:53:48] INFO: Waiting for application startup.\n",
"[2025-05-05 17:53:48] INFO: Application startup complete.\n",
"[2025-05-05 17:53:48] INFO: Uvicorn running on http://0.0.0.0:38475 (Press CTRL+C to quit)\n",
"[2025-05-05 17:53:48] INFO: 127.0.0.1:37502 - \"GET /v1/models HTTP/1.1\" 200 OK\n",
"[2025-05-05 17:53:49] INFO: 127.0.0.1:37516 - \"GET /get_model_info HTTP/1.1\" 200 OK\n",
"[2025-05-05 17:53:49] Prefill batch. #new-seq: 1, #new-token: 6, #cached-token: 0, token usage: 0.00, #running-req: 0, #queue-req: 0\n",
"2025-05-05 17:53:49,777 - INFO - flashinfer.jit: Loading JIT ops: batch_prefill_with_kv_cache_dtype_q_bf16_dtype_kv_bf16_dtype_o_bf16_dtype_idx_i32_head_dim_qk_128_head_dim_vo_128_posenc_0_use_swa_False_use_logits_cap_False_f16qk_False\n",
"2025-05-05 17:53:49,799 - INFO - flashinfer.jit: Finished loading JIT ops: batch_prefill_with_kv_cache_dtype_q_bf16_dtype_kv_bf16_dtype_o_bf16_dtype_idx_i32_head_dim_qk_128_head_dim_vo_128_posenc_0_use_swa_False_use_logits_cap_False_f16qk_False\n",
"[2025-05-05 17:53:50] INFO: 127.0.0.1:37526 - \"POST /generate HTTP/1.1\" 200 OK\n",
"[2025-05-05 17:53:50] The server is fired up and ready to roll!\n",
"\n",
"\n",
" NOTE: Typically, the server runs in a separate terminal.\n",
" In this notebook, we run the server and notebook code together, so their outputs are combined.\n",
" To improve clarity, the server logs are displayed in the original black color, while the notebook outputs are highlighted in blue.\n",
" We are running those notebooks in a CI parallel environment, so the throughput is not representative of the actual performance.\n",
" \n",
"Server started on http://localhost:38475\n"
]
}
],
"outputs": [],
"source": [
"from sglang import separate_reasoning, assistant_begin, assistant_end\n",
"from sglang import assistant, function, gen, system, user\n",
......@@ -105,15 +49,7 @@
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"[2025-05-05 17:53:53] INFO: 127.0.0.1:37530 - \"GET /get_model_info HTTP/1.1\" 200 OK\n"
]
}
],
"outputs": [],
"source": [
"set_default_backend(\n",
" RuntimeEndpoint(f\"http://localhost:{port}\", chat_template_name=\"qwen\")\n",
......@@ -131,41 +67,7 @@
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"[2025-05-05 17:53:53] Prefill batch. #new-seq: 1, #new-token: 31, #cached-token: 0, token usage: 0.00, #running-req: 0, #queue-req: 0\n",
"[2025-05-05 17:53:54] Decode batch. #running-req: 1, #token: 64, token usage: 0.00, gen throughput (token/s): 6.00, #queue-req: 0\n",
"[2025-05-05 17:53:54] Decode batch. #running-req: 1, #token: 104, token usage: 0.00, gen throughput (token/s): 82.06, #queue-req: 0\n",
"[2025-05-05 17:53:55] Decode batch. #running-req: 1, #token: 144, token usage: 0.00, gen throughput (token/s): 81.56, #queue-req: 0\n",
"[2025-05-05 17:53:55] Decode batch. #running-req: 1, #token: 184, token usage: 0.00, gen throughput (token/s): 81.14, #queue-req: 0\n",
"[2025-05-05 17:53:56] Decode batch. #running-req: 1, #token: 224, token usage: 0.00, gen throughput (token/s): 80.91, #queue-req: 0\n",
"[2025-05-05 17:53:56] Decode batch. #running-req: 1, #token: 264, token usage: 0.00, gen throughput (token/s): 80.55, #queue-req: 0\n",
"[2025-05-05 17:53:56] INFO: 127.0.0.1:37538 - \"POST /generate HTTP/1.1\" 200 OK\n",
"<think>\n",
"Okay, the user is asking for three countries and their capitals. Let me think about which countries to choose. I should pick some well-known ones to make it easy for the user.\n",
"\n",
"First, France is a good start because its capital is Paris, which is a major city. Then maybe Germany with Berlin. Those are both in Europe and have clear capitals. \n",
"\n",
"Next, I need a country from another continent. Let's go with Japan, which has Tokyo as its capital. That covers Asia. \n",
"\n",
"Wait, should I check if there are any countries with non-obvious capitals? Maybe not necessary. The user probably wants straightforward answers. \n",
"\n",
"Let me confirm the capitals again. France - Paris, Germany - Berlin, Japan - Tokyo. Yep, that's correct. \n",
"\n",
"I should present them in a clear list. Maybe number them and list each with the capital. Keep it simple and to the point. No need for extra info unless the user asks. \n",
"\n",
"Alright, that should cover it. Three countries, their capitals, correct and easy to understand.\n",
"</think>\n",
"\n",
"1. **France** - Paris \n",
"2. **Germany** - Berlin \n",
"3. **Japan** - Tokyo\n"
]
}
],
"outputs": [],
"source": [
"@function\n",
"def basic_qa(s, question):\n",
......@@ -191,38 +93,7 @@
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"dict_keys(['answer', 'answer_reasoning_content'])\n",
"[2025-05-05 17:56:44] Prefill batch. #new-seq: 1, #new-token: 1, #cached-token: 30, token usage: 0.00, #running-req: 0, #queue-req: 0\n",
"[2025-05-05 17:56:44] Decode batch. #running-req: 1, #token: 63, token usage: 0.00, gen throughput (token/s): 3.77, #queue-req: 0\n",
"[2025-05-05 17:56:45] Decode batch. #running-req: 1, #token: 103, token usage: 0.00, gen throughput (token/s): 82.12, #queue-req: 0\n",
"[2025-05-05 17:56:45] Decode batch. #running-req: 1, #token: 143, token usage: 0.00, gen throughput (token/s): 81.60, #queue-req: 0\n",
"[2025-05-05 17:56:46] Decode batch. #running-req: 1, #token: 183, token usage: 0.00, gen throughput (token/s): 81.17, #queue-req: 0\n",
"[2025-05-05 17:56:46] Decode batch. #running-req: 1, #token: 223, token usage: 0.00, gen throughput (token/s): 80.90, #queue-req: 0\n",
"[2025-05-05 17:56:46] INFO: 127.0.0.1:45282 - \"POST /generate HTTP/1.1\" 200 OK\n",
"\n",
"Separated Reasoning Content:\n",
"Okay, the user is asking for three countries and their capitals. Let me think. I need to make sure the countries are correct and their capitals are properly matched.\n",
"\n",
"First, I should start with a well-known country. France is a good example. Its capital is Paris. That's straightforward. Next, maybe a country in Asia. Japan's capital is Tokyo. That's correct. Then, perhaps a country in Africa. Egypt's capital is Cairo. Wait, is that right? Yes, Egypt's capital is indeed Cairo. Let me double-check. France - Paris, Japan - Tokyo, Egypt - Cairo. Those are all correct. I should present them in a clear list format. Make sure the country names are spelled correctly and the capitals are properly capitalized. No need for any extra information, just the three pairs. That should answer the user's question effectively.\n",
"\n",
"\n",
"\n",
"Content:\n",
"1. **France** - Paris \n",
"2. **Japan** - Tokyo \n",
"3. **Egypt** - Cairo\n",
"\n",
"\n",
"Messages:\n",
"{'role': 'assistant', 'content': '1. **France** - Paris \\n2. **Japan** - Tokyo \\n3. **Egypt** - Cairo'}\n"
]
}
],
"outputs": [],
"source": [
"@function\n",
"def basic_qa_separate_reasoning(s, question):\n",
......@@ -254,71 +125,7 @@
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"[2025-05-05 17:54:03] Decode batch. #running-req: 1, #token: 0, token usage: 0.00, gen throughput (token/s): 79.25, #queue-req: 0\n",
"[2025-05-05 17:54:03] Prefill batch. #new-seq: 1, #new-token: 18, #cached-token: 18, token usage: 0.00, #running-req: 0, #queue-req: 0\n",
"[2025-05-05 17:54:03] Decode batch. #running-req: 1, #token: 77, token usage: 0.00, gen throughput (token/s): 75.90, #queue-req: 0\n",
"[2025-05-05 17:54:04] Decode batch. #running-req: 1, #token: 117, token usage: 0.00, gen throughput (token/s): 81.85, #queue-req: 0\n",
"[2025-05-05 17:54:04] Decode batch. #running-req: 1, #token: 157, token usage: 0.00, gen throughput (token/s): 81.36, #queue-req: 0\n",
"[2025-05-05 17:54:05] Decode batch. #running-req: 1, #token: 197, token usage: 0.00, gen throughput (token/s): 81.01, #queue-req: 0\n",
"[2025-05-05 17:54:05] Decode batch. #running-req: 1, #token: 237, token usage: 0.00, gen throughput (token/s): 80.80, #queue-req: 0\n",
"[2025-05-05 17:54:06] Decode batch. #running-req: 1, #token: 277, token usage: 0.00, gen throughput (token/s): 80.43, #queue-req: 0\n",
"[2025-05-05 17:54:06] Decode batch. #running-req: 1, #token: 317, token usage: 0.00, gen throughput (token/s): 80.10, #queue-req: 0\n",
"[2025-05-05 17:54:07] Decode batch. #running-req: 1, #token: 357, token usage: 0.00, gen throughput (token/s): 79.83, #queue-req: 0\n",
"[2025-05-05 17:54:07] INFO: 127.0.0.1:41424 - \"POST /generate HTTP/1.1\" 200 OK\n",
"\n",
"\n",
"first_answer:\n",
"Here’s a list of three countries and their capitals:\n",
"\n",
"1. **France** – **Paris** \n",
"2. **United States** – **Washington, D.C.** \n",
"3. **Brazil** – **Brasília** \n",
"\n",
"Let me know if you'd like more examples! 😊\n",
"\n",
"\n",
"first_answer_reasoning_content:\n",
"Okay, the user is asking for a list of three countries and their capitals. Let me think about which countries to choose. They might be a student studying geography or someone just curious. I should pick well-known countries to make it easier for them.\n",
"\n",
"First, I'll start with the most obvious ones. France and its capital Paris are a classic example. Then, maybe the United States with Washington, D.C. That's another common one. For the third country, perhaps Brazil with Brasília? Wait, I should make sure I'm correct about the capitals. Let me double-check: France is Paris, USA is Washington, D.C., and Brazil is indeed Brasília. \n",
"\n",
"Alternatively, maybe including a country from a different continent could be better? Like Japan with Tokyo? But the user didn't specify any particular region. Since the first two are from Europe and North America, adding a South American country might be a good mix. \n",
"\n",
"Wait, but the user just asked for three, so as long as they're accurate, it's fine. I'll go with France, USA, and Brazil. Let me make sure I get the spelling right. Paris, Washington D.C., Brasília. Yeah, that's correct. I should present them in a clear list format. The user might need this for a school assignment or a quiz. Alright, that should cover it.\n",
"\n",
"[2025-05-05 17:54:07] Prefill batch. #new-seq: 1, #new-token: 83, #cached-token: 36, token usage: 0.00, #running-req: 0, #queue-req: 0\n",
"[2025-05-05 17:54:07] Decode batch. #running-req: 1, #token: 138, token usage: 0.00, gen throughput (token/s): 76.16, #queue-req: 0\n",
"[2025-05-05 17:54:08] Decode batch. #running-req: 1, #token: 178, token usage: 0.00, gen throughput (token/s): 81.10, #queue-req: 0\n",
"[2025-05-05 17:54:08] Decode batch. #running-req: 1, #token: 218, token usage: 0.00, gen throughput (token/s): 80.91, #queue-req: 0\n",
"[2025-05-05 17:54:09] Decode batch. #running-req: 1, #token: 258, token usage: 0.00, gen throughput (token/s): 80.63, #queue-req: 0\n",
"[2025-05-05 17:54:09] Decode batch. #running-req: 1, #token: 298, token usage: 0.00, gen throughput (token/s): 80.29, #queue-req: 0\n",
"[2025-05-05 17:54:10] Decode batch. #running-req: 1, #token: 338, token usage: 0.00, gen throughput (token/s): 79.96, #queue-req: 0\n",
"[2025-05-05 17:54:10] INFO: 127.0.0.1:47266 - \"POST /generate HTTP/1.1\" 200 OK\n",
"\n",
"\n",
"second_answer:\n",
"Here’s another list of three countries and their capitals:\n",
"\n",
"1. **Nigeria** – **Lagos** \n",
"2. **Japan** – **Tokyo** \n",
"3. **Argentina** – **Buenos Aires** \n",
"\n",
"Let me know if you'd like more examples! 😊\n",
"\n",
"\n",
"second_answer_reasoning_content:\n",
"Okay, the user asked for another list of three countries and their capitals. Let me think about what they might need. They previously got France, the US, and Brazil. Maybe they want more variety or different regions? I should pick countries from different continents to cover a broad range.\n",
"\n",
"First, maybe include a country from Africa. Lagos is the capital of Nigeria, which is a common example. Then, Asia – maybe Japan, with Tokyo. That's a major country. Then, a country from South America, like Argentina with Buenos Aires. That gives a good mix. I should check if those capitals are correct. Lagos is right for Nigeria, Tokyo for Japan, and Buenos Aires for Argentina. Yeah, that works. I'll present them in a list format again, making sure to mention each country and its capital clearly. Make sure the response is friendly and offers further help if needed.\n",
"\n"
]
}
],
"outputs": [],
"source": [
"@function\n",
"def multi_turn_qa(s):\n",
......@@ -360,23 +167,7 @@
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"[2025-05-05 17:54:10] Prefill batch. #new-seq: 1, #new-token: 9, #cached-token: 26, token usage: 0.00, #running-req: 0, #queue-req: 0\n",
"[2025-05-05 17:54:10] Decode batch. #running-req: 1, #token: 51, token usage: 0.00, gen throughput (token/s): 76.50, #queue-req: 0\n",
"[2025-05-05 17:54:10] INFO: 127.0.0.1:47276 - \"POST /generate HTTP/1.1\" 200 OK\n",
"Reasoning Content:\n",
" \n",
"Content:\n",
" 1. France - Paris \n",
"2. Germany - Berlin \n",
"3. Japan - Tokyo\n"
]
}
],
"outputs": [],
"source": [
"reasoning_state = basic_qa_separate_reasoning(\n",
" \"List 3 countries and their capitals. /no_think\"\n",
......@@ -423,37 +214,7 @@
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"[2025-05-05 17:54:11] Prefill batch. #new-seq: 1, #new-token: 26, #cached-token: 8, token usage: 0.00, #running-req: 0, #queue-req: 0\n",
"[2025-05-05 17:54:11] Decode batch. #running-req: 1, #token: 68, token usage: 0.00, gen throughput (token/s): 47.33, #queue-req: 0\n",
"[2025-05-05 17:54:12] Decode batch. #running-req: 1, #token: 108, token usage: 0.00, gen throughput (token/s): 83.03, #queue-req: 0\n",
"[2025-05-05 17:54:12] Decode batch. #running-req: 1, #token: 148, token usage: 0.00, gen throughput (token/s): 82.51, #queue-req: 0\n",
"[2025-05-05 17:54:13] Decode batch. #running-req: 1, #token: 188, token usage: 0.00, gen throughput (token/s): 82.06, #queue-req: 0\n",
"[2025-05-05 17:54:13] Decode batch. #running-req: 1, #token: 228, token usage: 0.00, gen throughput (token/s): 81.80, #queue-req: 0\n",
"[2025-05-05 17:54:14] Decode batch. #running-req: 1, #token: 268, token usage: 0.00, gen throughput (token/s): 81.48, #queue-req: 0\n",
"[2025-05-05 17:54:14] Decode batch. #running-req: 1, #token: 308, token usage: 0.00, gen throughput (token/s): 81.14, #queue-req: 0\n",
"[2025-05-05 17:54:15] Decode batch. #running-req: 1, #token: 348, token usage: 0.00, gen throughput (token/s): 80.84, #queue-req: 0\n",
"[2025-05-05 17:54:15] INFO: 127.0.0.1:47290 - \"POST /generate HTTP/1.1\" 200 OK\n",
"Answer:\n",
"2023-10-05\n",
"\n",
"\n",
"Reasoning Content:\n",
"Okay, the user is asking for the IP addresses of Google's DNS servers. Let me recall what I know about DNS servers. Google provides two public DNS servers, right? They're commonly used for their reliability and speed.\n",
"\n",
"I think the primary one is 8.8.8.8. Wait, isn't there another one? Oh yeah, 8.8.4.4. Those are the two main ones. Let me make sure I'm not mixing them up with other providers. For example, Cloudflare uses 1.1.1.1 and 1.0.0.1. But Google's are definitely 8.8.8.8 and 8.8.4.4. \n",
"\n",
"I should check if there are any other IP addresses, but I don't think so. They have two main ones. The user might be looking to set up their DNS settings, so providing both is important. Also, maybe mention that they're both in the same range, which is 8.8.0.0/14. But the user just asked for the IP addresses, so maybe just list them. \n",
"\n",
"Wait, the user said \"just provide the answer,\" so maybe they don't need extra info. But to be thorough, I should confirm that those are the correct ones. Let me think if there's any chance of confusion. No, 8.8.8.8 is the primary, and 8.8.4.4 is the secondary. Yeah, that's right. So the answer is those two IPs.\n",
"\n"
]
}
],
"outputs": [],
"source": [
"print_highlight(f\"Answer:\\n{reasoning_state['answer']}\")\n",
"print_highlight(\n",
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment