Krishna3451112/cloud-sec
Text Generation • Updated
messages listlengths 37 61 | task_id stringclasses 1
value | model_source stringclasses 2
values | terminal_reward float64 0.85 1 | total_reward float64 1.15 1.5 | num_steps int64 18 30 |
|---|---|---|---|---|---|
[
{
"role": "system",
"content": "You are an on-call Site Reliability Engineer (SRE) at NimbusGuard, a cloud-security SaaS.\n\nYou've been paged with an incident alert. Investigate using the tools available, identify the root cause, and propose a fix.\n\n**Environment:** 3 cloud deployments (cloud-1 us-east, ... | task_01_oidc_rotation | claude-opus-4-5 | 1 | 1.4 | 21 |
[
{
"role": "system",
"content": "You are an on-call Site Reliability Engineer (SRE) at NimbusGuard, a cloud-security SaaS.\n\nYou've been paged with an incident alert. Investigate using the tools available, identify the root cause, and propose a fix.\n\n**Environment:** 3 cloud deployments (cloud-1 us-east, ... | task_01_oidc_rotation | claude-opus-4-5 | 1 | 1.5 | 22 |
[
{
"role": "system",
"content": "You are an on-call Site Reliability Engineer (SRE) at NimbusGuard, a cloud-security SaaS.\n\nYou've been paged with an incident alert. Investigate using the tools available, identify the root cause, and propose a fix.\n\n**Environment:** 3 cloud deployments (cloud-1 us-east, ... | task_01_oidc_rotation | claude-opus-4-5 | 1 | 1.5 | 21 |
[
{
"role": "system",
"content": "You are an on-call Site Reliability Engineer (SRE) at NimbusGuard, a cloud-security SaaS.\n\nYou've been paged with an incident alert. Investigate using the tools available, identify the root cause, and propose a fix.\n\n**Environment:** 3 cloud deployments (cloud-1 us-east, ... | task_01_oidc_rotation | claude-opus-4-5 | 1 | 1.5 | 18 |
[
{
"role": "system",
"content": "You are an on-call Site Reliability Engineer (SRE) at NimbusGuard, a cloud-security SaaS.\n\nYou've been paged with an incident alert. Investigate using the tools available, identify the root cause, and propose a fix.\n\n**Environment:** 3 cloud deployments (cloud-1 us-east, ... | task_01_oidc_rotation | claude-opus-4-5 | 1 | 1.5 | 18 |
[
{
"role": "system",
"content": "You are an on-call Site Reliability Engineer (SRE) at NimbusGuard, a cloud-security SaaS.\n\nYou've been paged with an incident alert. Investigate using the tools available, identify the root cause, and propose a fix.\n\n**Environment:** 3 cloud deployments (cloud-1 us-east, ... | task_01_oidc_rotation | claude-opus-4-5 | 1 | 1.5 | 22 |
[{"role":"system","content":"You are an on-call Site Reliability Engineer (SRE) at NimbusGuard, a cl(...TRUNCATED) | task_01_oidc_rotation | claude-opus-4-5 | 1 | 1.4 | 20 |
[{"role":"system","content":"You are an on-call Site Reliability Engineer (SRE) at NimbusGuard, a cl(...TRUNCATED) | task_01_oidc_rotation | claude-opus-4-5 | 1 | 1.5 | 21 |
[{"role":"system","content":"You are an on-call Site Reliability Engineer (SRE) at NimbusGuard, a cl(...TRUNCATED) | task_01_oidc_rotation | claude-opus-4-5 | 1 | 1.5 | 23 |
[{"role":"system","content":"You are an on-call Site Reliability Engineer (SRE) at NimbusGuard, a cl(...TRUNCATED) | task_01_oidc_rotation | claude-opus-4-5 | 1 | 1.5 | 23 |
YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
Opus-generated trajectories for fine-tuning a small LLM (Qwen2.5-7B) to investigate cloud-security incidents in our Cloud Sec Env.
Each row is one full trajectory (system prompt + alert + alternating tool calls and
results, ending with a submit_answer action). Assistant turns are pre-formatted as
JSON objects of the shape {"reasoning", "tool_name", "arguments"} so a fine-tune
on this data produces parseable JSON output end-to-end.
Filtered for terminal_reward >= 0.5 under our deterministic keyword rubric.
from datasets import load_dataset
ds = load_dataset("Krishna3451112/cloud-sec-env-sft", split="train")