@delta-infra/cli 1.1.17 → 1.1.18

package/skills/delta-science/references/synbo.md

@@ -1,198 +1,201 @@
-# SynBO operation 参考
-
-所有在线调用都通过 `delta-cli science invoke`，服务地址来自 CLI 的标准
-`base_url` 派生配置（`{base_url}/science_tool`）。这里的路径只说明业务 operation，不得据此直连 HTTP。
-
-用户说“优化反应条件”“推荐下一批实验”“偶联反应怎么继续做”，或给出溶剂、温度、
-催化剂、碱、配体和历史得率时，即使没有提到 SynBO、BO、CLI 或字段名也使用本节。
-每个用户任务最多一次 SynBO 业务调用；成功后立即结束，validation、502、504、
-timeout、断连或空 recommendations 均不得换参数、换模型或重试。
-
-## Health
-
-operation：`health`，无 body。
-
-Reports SynBO import status, Torch/CUDA runtime availability, service limits,
-default device, and visible CUDA devices.
-
-Important fields inside gateway `data`:
-
-```json
-{
-  "status": "ok",
-  "synbo_available": true,
-  "metadata": {
-    "service": "synbo-service",
-    "engine": "synbo",
-    "version": "0.1.0",
-    "synbo_version": "0.1.0",
-    "default_device": "cpu",
-    "cuda_visible_devices": ""
-  },
-  "limits": {
-    "max_batch_size": 128,
-    "max_reaction_space_size": 200000,
-    "max_previous_rows": 10000
-  },
-  "runtime": {
-    "torch_available": true,
-    "cuda_available": false,
-    "cuda_device_count": 0,
-    "cuda_device_names": []
-  },
-  "warnings": []
-}
-```
-
-## Initialize
-
-operation：`initialize`。
-
-Use for initial condition sampling when no measured previous results exist.
-
-Request body:
-
-```json
-{
-  "condition_dict": {
-    "catalyst": ["Pd(OAc)2", "Pd(PPh3)4", "Pd2(dba)3"],
-    "solvent": ["THF", "Dioxane", "Toluene"],
-    "base": ["K2CO3", "NaOEt", "DBU"]
-  },
-  "opt_metrics": ["yield"],
-  "opt_metric_settings": [
-    {"opt_direct": "max", "opt_range": [0, 100], "metric_weight": 1.0}
-  ],
-  "batch_size": 3,
-  "sampling_method": "random",
-  "desc_normalize": "minmax",
-  "random_seed": 42
-}
-```
-
-Supported `sampling_method` values: `sobol`, `random`, `lhs`, `kmeans`.
-
-## Optimize
-
-operation：`optimize`。
-
-Use after measured experiment rows are available. Each previous row must contain
-all condition columns and all objective metric columns.
-
-Request body:
-
-```json
-{
-  "condition_dict": {
-    "catalyst": ["Pd(OAc)2", "Pd(PPh3)4", "Pd2(dba)3"],
-    "solvent": ["THF", "Dioxane", "Toluene"],
-    "base": ["K2CO3", "NaOEt", "DBU"]
-  },
-  "opt_metrics": ["yield"],
-  "previous_results": [
-    {"batch": 0, "catalyst": "Pd(OAc)2", "solvent": "THF", "base": "K2CO3", "yield": 42.0},
-    {"batch": 0, "catalyst": "Pd(PPh3)4", "solvent": "Dioxane", "base": "NaOEt", "yield": 55.0},
-    {"batch": 0, "catalyst": "Pd2(dba)3", "solvent": "Toluene", "base": "DBU", "yield": 63.0}
-  ],
-  "batch_size": 2,
-  "accuracy": "tiny",
-  "surrogate_model": "RF",
-  "acq_func": "UCB",
-  "device": "cpu",
-  "random_seed": 42
-}
-```
-
-### 数值条件必须先离散化
-
-SynBO 的 `condition_dict` 只接受显式列表，不接受连续范围对象。对于“温度
-40–100°C、催化剂用量 0.5–2.0%”并已有 60/70/80/90°C 与
-0.5/1.0/1.5/2.0% 历史值的任务，使用：
-
-```json
-{
-  "condition_dict": {
-    "solvent": ["ethanol", "acetonitrile", "DMSO"],
-    "temperature_C": ["40", "60", "70", "80", "90", "100"],
-    "catalyst_loading_pct": ["0.5", "1.0", "1.5", "2.0"]
-  },
-  "opt_metrics": ["yield"],
-  "previous_results": [
-    {
-      "batch": 0,
-      "solvent": "ethanol",
-      "temperature_C": "60",
-      "catalyst_loading_pct": "1.0",
-      "yield": 55.0
-    }
-  ],
-  "batch_size": 3,
-  "accuracy": "tiny",
-  "surrogate_model": "RF",
-  "acq_func": "UCB",
-  "device": "cpu",
-  "random_seed": 42
-}
-```
-
-约束：
-
-- 每个条件值统一为字符串，objective metric 保持数值。
-- 每条 `previous_results` 中的条件值必须出现在对应 `condition_dict` 列表中。
-- 连续范围的离散网格必须包含用户上下界和全部历史观测值；最终答案必须说明实际网格。
-- 将条件值显式序列化为字符串，并确保每条历史记录只使用 `condition_dict` 中列出的值。
-  如需小数据 CPU 设置，显式传递 `accuracy:"tiny"`、`surrogate_model:"RF"`、
-  `acq_func:"UCB"` 和 `device:"cpu"`；不得假定客户端会补充默认值。
-
-Useful optional fields:
-
-- `accuracy`: `tiny`, `low`, `medium`, `high`, or `ultra`.
-- `acq_func`: `EHVI`, `UCB`, `ParEGO`, or `NEI`.
-- `surrogate_model`: `GP`, `RF`, `BNN`, or `BayesianLinear`.
-- `device`: `cpu`, `auto`, `cuda`, or `cuda:0`.
-- `cuda_visible_devices`: request-level CUDA visibility hint. Use only when the
-  user explicitly asks for it and `/health` indicates a CUDA-capable runtime.
-- `constraints`: map condition type to prohibited values.
-
-## Descriptors
-
-If omitted, SynBO uses generated OneHot descriptors. To provide descriptors:
-
-```json
-{
-  "descriptors": {
-    "solvent": [
-      {"name": "THF", "polarity": 0.207, "boiling_point_C": 66.0},
-      {"name": "Toluene", "polarity": 0.099, "boiling_point_C": 110.6}
-    ]
-  },
-  "descriptor_index_col": "name"
-}
-```
-
-`descriptors` must contain exactly the same keys as `condition_dict`. Descriptor
-rows must include the configured index column and numeric descriptor columns.
-
-## Error Handling
-
-Gateway errors have the standard envelope:
-
-```json
-{
-  "code": 1,
-  "message": "SynBO service returned HTTP 400",
-  "data": {
-    "endpoint": "/synbo/optimize",
-    "synbo_response": {
-      "detail": "previous_results is missing metric columns: ['yield']"
-    }
-  }
-}
-```
-
-常见原因包括条件空间过大、条件值重复、历史条件值不在 condition_dict 中、
-descriptor 非数值、历史行缺列、metric 非数值，以及无 GPU 时请求 CUDA。
-
-服务端有独立于客户端的单次请求时限。GP、EHVI、高 accuracy 或共享资源繁忙可能导致
-timeout。普通用户任务中不得自动换参数重试；应报告“未产生推荐”，保留本次错误，
-由用户决定是否另起一次明确的回归测试。
+# SynBO operation reference
+
+Make all online calls with `delta-cli science invoke`; the service address comes from the CLI's
+standard `base_url` configuration (`{base_url}/science_tool`). The paths here describe business
+operations only and do not authorise direct HTTP access.
+
+Use this section when the user asks to optimise reaction conditions, recommend the next batch of
+experiments, continue a coupling reaction, or provides solvents, temperatures, catalysts, bases,
+ligands and historical yields—even if they do not mention SynBO, BO, the CLI or field names. Make
+at most one SynBO business call per user task. Stop immediately after success; for validation
+errors, 502, 504, timeouts, disconnections or empty recommendations, do not change parameters or
+models, or retry.
+
+## Health
+
+Operation: `health`; no request body.
+
+Reports SynBO import status, Torch/CUDA runtime availability, service limits,
+default device and visible CUDA devices.
+
+Important fields inside gateway `data`:
+
+```json
+{
+  "status": "ok",
+  "synbo_available": true,
+  "metadata": {
+    "service": "synbo-service",
+    "engine": "synbo",
+    "version": "0.1.0",
+    "synbo_version": "0.1.0",
+    "default_device": "cpu",
+    "cuda_visible_devices": ""
+  },
+  "limits": {
+    "max_batch_size": 128,
+    "max_reaction_space_size": 200000,
+    "max_previous_rows": 10000
+  },
+  "runtime": {
+    "torch_available": true,
+    "cuda_available": false,
+    "cuda_device_count": 0,
+    "cuda_device_names": []
+  },
+  "warnings": []
+}
+```
+
+## Initialize
+
+Operation: `initialize`.
+
+Use for initial condition sampling when no measured previous results exist.
+
+Request body:
+
+```json
+{
+  "condition_dict": {
+    "catalyst": ["Pd(OAc)2", "Pd(PPh3)4", "Pd2(dba)3"],
+    "solvent": ["THF", "Dioxane", "Toluene"],
+    "base": ["K2CO3", "NaOEt", "DBU"]
+  },
+  "opt_metrics": ["yield"],
+  "opt_metric_settings": [
+    {"opt_direct": "max", "opt_range": [0, 100], "metric_weight": 1.0}
+  ],
+  "batch_size": 3,
+  "sampling_method": "random",
+  "desc_normalize": "minmax",
+  "random_seed": 42
+}
+```
+
+Supported `sampling_method` values: `sobol`, `random`, `lhs`, `kmeans`.
+
+## Optimize
+
+Operation: `optimize`.
+
+Use after measured experiment rows are available. Each previous row must contain
+all condition columns and all objective metric columns.
+
+Request body:
+
+```json
+{
+  "condition_dict": {
+    "catalyst": ["Pd(OAc)2", "Pd(PPh3)4", "Pd2(dba)3"],
+    "solvent": ["THF", "Dioxane", "Toluene"],
+    "base": ["K2CO3", "NaOEt", "DBU"]
+  },
+  "opt_metrics": ["yield"],
+  "previous_results": [
+    {"batch": 0, "catalyst": "Pd(OAc)2", "solvent": "THF", "base": "K2CO3", "yield": 42.0},
+    {"batch": 0, "catalyst": "Pd(PPh3)4", "solvent": "Dioxane", "base": "NaOEt", "yield": 55.0},
+    {"batch": 0, "catalyst": "Pd2(dba)3", "solvent": "Toluene", "base": "DBU", "yield": 63.0}
+  ],
+  "batch_size": 2,
+  "accuracy": "tiny",
+  "surrogate_model": "RF",
+  "acq_func": "UCB",
+  "device": "cpu",
+  "random_seed": 42
+}
+```
+
+### Discretise numeric conditions first
+
+SynBO's `condition_dict` accepts explicit lists only; it does not accept continuous range
+objects. For a task with a temperature range of 40–100°C and catalyst loading of 0.5–2.0%, with
+historical values at 60/70/80/90°C and 0.5/1.0/1.5/2.0%, use:
+
+```json
+{
+  "condition_dict": {
+    "solvent": ["ethanol", "acetonitrile", "DMSO"],
+    "temperature_C": ["40", "60", "70", "80", "90", "100"],
+    "catalyst_loading_pct": ["0.5", "1.0", "1.5", "2.0"]
+  },
+  "opt_metrics": ["yield"],
+  "previous_results": [
+    {
+      "batch": 0,
+      "solvent": "ethanol",
+      "temperature_C": "60",
+      "catalyst_loading_pct": "1.0",
+      "yield": 55.0
+    }
+  ],
+  "batch_size": 3,
+  "accuracy": "tiny",
+  "surrogate_model": "RF",
+  "acq_func": "UCB",
+  "device": "cpu",
+  "random_seed": 42
+}
+```
+
+Constraints:
+
+- Represent every condition value as a string and keep objective metrics numeric.
+- Every condition value in each `previous_results` row must appear in the corresponding `condition_dict` list.
+- The discretised grid for a continuous range must include the user's lower and upper bounds and all historical observations. State the actual grid in the final answer.
+- Serialise condition values explicitly as strings, and ensure each historical row uses only values listed in `condition_dict`. For a small-data CPU configuration, pass `accuracy:"tiny"`, `surrogate_model:"RF"`, `acq_func:"UCB"` and `device:"cpu"` explicitly; do not assume the client supplies defaults.
+
+Useful optional fields:
+
+- `accuracy`: `tiny`, `low`, `medium`, `high`, or `ultra`.
+- `acq_func`: `EHVI`, `UCB`, `ParEGO`, or `NEI`.
+- `surrogate_model`: `GP`, `RF`, `BNN`, or `BayesianLinear`.
+- `device`: `cpu`, `auto`, `cuda`, or `cuda:0`.
+- `cuda_visible_devices`: request-level CUDA visibility hint. Use only when the
+  user explicitly asks for it and `/health` indicates a CUDA-capable runtime.
+- `constraints`: map condition type to prohibited values.
+
+## Descriptors
+
+If omitted, SynBO uses generated OneHot descriptors. To provide descriptors:
+
+```json
+{
+  "descriptors": {
+    "solvent": [
+      {"name": "THF", "polarity": 0.207, "boiling_point_C": 66.0},
+      {"name": "Toluene", "polarity": 0.099, "boiling_point_C": 110.6}
+    ]
+  },
+  "descriptor_index_col": "name"
+}
+```
+
+`descriptors` must contain exactly the same keys as `condition_dict`. Descriptor
+rows must include the configured index column and numeric descriptor columns.
+
+## Error Handling
+
+Gateway errors have the standard envelope:
+
+```json
+{
+  "code": 1,
+  "message": "SynBO service returned HTTP 400",
+  "data": {
+    "endpoint": "/synbo/optimize",
+    "synbo_response": {
+      "detail": "previous_results is missing metric columns: ['yield']"
+    }
+  }
+}
+```
+
+Common causes include an oversized condition space, duplicate condition values, historical
+condition values absent from `condition_dict`, non-numeric descriptors, missing columns in
+historical rows, non-numeric metrics, and a CUDA request when no GPU is available.
+
+The server has a per-request timeout independent of the client. GP, EHVI, high accuracy or busy
+shared resources may cause a timeout. Do not automatically change parameters and retry during an
+ordinary user task. Report “no recommendations were produced”, preserve the error, and let the
+user decide whether to start a separate, explicitly requested regression test.