Offlin33er commited on
Commit
c5ca5ac
·
verified ·
1 Parent(s): c75d52d

Cross-link Oxford Digital 6,000-row v4 Enterprise Security Suite

Browse files
Files changed (1) hide show
  1. README.md +49 -43
README.md CHANGED
@@ -1,44 +1,50 @@
1
- ---
2
- language:
3
- - en
4
- task_categories:
5
- - text-generation
6
- tags:
7
- - function-calling
8
- - tool-use
9
- - synthetic
10
- - glm
11
- - conversational
12
- size_categories:
13
- - n<1K
14
- ---
15
-
16
- # GLM-5.3-Flash function-calling (synthetic)
17
-
18
- 500 synthetic function-calling training samples generated with
19
- [zai-org/GLM-5.3-Flash](https://huggingface.co/zai-org/GLM-5.3-Flash) via HF Inference
20
- Providers (auto-routing with novita / together / fireworks fallback), 2026-09-21.
21
-
22
- ## Schema
23
-
24
- Each row:
25
-
26
- - `messages` — chat in OpenAI tool-use format (`system` / `user` / `assistant`); the final
27
- assistant message either carries `tool_calls` (with JSON-string `arguments`) or is a plain-text answer
28
- - `tools` — 1–4 tool schemas in OpenAI function format (`type`, `function{name, description, parameters}`)
29
- - `category` — one of:
30
- - `single_call` — exactly one tool call
31
- - `parallel_calls` — 2–4 independent calls in one response
32
- - `multi_turn` — multi-turn with tool-result turns
33
- - `error_handling` — call with plausibly failing arguments
34
- - `no_tool` — tempting but irrelevant tools; assistant answers directly
35
-
36
- Domains span weather, travel, finance, e-commerce, smart home, file management, and ~14 more.
37
- Every sample passed a validator: tool-call names resolve to declared tools, `arguments` parse as
38
- JSON objects, last message is assistant-role, message roles valid.
39
-
40
- ## Generation
41
-
42
- Single-shot JSON generation per sample (`temperature=0.9`, seeded), 10-way concurrency,
43
- per-sample validation with provider fallback. Generation script pattern available in the
 
 
 
 
 
 
44
  conversation that produced this dataset. Costs: ~7 min on `cpu-basic` (Jobs) + provider inference.
 
1
+ ---
2
+ language:
3
+ - en
4
+ task_categories:
5
+ - text-generation
6
+ tags:
7
+ - function-calling
8
+ - tool-use
9
+ - synthetic
10
+ - glm
11
+ - conversational
12
+ size_categories:
13
+ - n<1K
14
+ ---
15
+
16
+ # GLM-5.3-Flash function-calling (synthetic)
17
+
18
+ <!-- OXFORD_DIGITAL_V4_BANNER_START -->
19
+ > **Commercial 6,000-Row Multi-Domain Security Suite v4 (Zero Overlap with v1–v3):** Need an unseen, contamination-free benchmark to evaluate or fine-tune beyond the public v1–v3 splits? Explore the **6,000-row Oxford Digital Enterprise Security Remediation Suite v4** (`1,200` Solidity EVM + `1,200` Rust/Solana Anchor + `1,200` Python/TS AI Agent & MCP + `1,200` Go/TS/Python Cloud APIs + `1,200` K8s/Terraform IaC in Parquet/CSV/JSONL):
20
+ > - **Interactive 100-Row Free Sample:** [`Offlin33er/solidity-exploit-reasoning-v4-sample`](https://huggingface.co/datasets/Offlin33er/solidity-exploit-reasoning-v4-sample)
21
+ > - **Instant Commercial License ($599 Self-Serve Pro / $1,499 Enterprise):** [Get Full 6,000-Row v4 Suite](https://buy.polar.sh/polar_cl_BRF5b1REPnB25KchbKx85ZckWVA5ITZf1nkVi3YVYn8) · Direct Enterprise / Instant USDC: `rayxxxnd@oxforddigitals.com`
22
+ <!-- OXFORD_DIGITAL_V4_BANNER_END -->
23
+
24
+ 500 synthetic function-calling training samples generated with
25
+ [zai-org/GLM-5.3-Flash](https://huggingface.co/zai-org/GLM-5.3-Flash) via HF Inference
26
+ Providers (auto-routing with novita / together / fireworks fallback), 2026-09-21.
27
+
28
+ ## Schema
29
+
30
+ Each row:
31
+
32
+ - `messages` — chat in OpenAI tool-use format (`system` / `user` / `assistant`); the final
33
+ assistant message either carries `tool_calls` (with JSON-string `arguments`) or is a plain-text answer
34
+ - `tools` — 1–4 tool schemas in OpenAI function format (`type`, `function{name, description, parameters}`)
35
+ - `category` — one of:
36
+ - `single_call` — exactly one tool call
37
+ - `parallel_calls` — 2–4 independent calls in one response
38
+ - `multi_turn` — multi-turn with tool-result turns
39
+ - `error_handling` — call with plausibly failing arguments
40
+ - `no_tool` — tempting but irrelevant tools; assistant answers directly
41
+
42
+ Domains span weather, travel, finance, e-commerce, smart home, file management, and ~14 more.
43
+ Every sample passed a validator: tool-call names resolve to declared tools, `arguments` parse as
44
+ JSON objects, last message is assistant-role, message roles valid.
45
+
46
+ ## Generation
47
+
48
+ Single-shot JSON generation per sample (`temperature=0.9`, seeded), 10-way concurrency,
49
+ per-sample validation with provider fallback. Generation script pattern available in the
50
  conversation that produced this dataset. Costs: ~7 min on `cpu-basic` (Jobs) + provider inference.