Dataset Viewer
Auto-converted to Parquet Duplicate
messages
list
study
string
dataset
string
split
string
n_variables
int64
provenance
string
[ { "role": "system", "content": "You are a clinical SAS programming assistant." }, { "role": "user", "content": "Write SAS code to derive and build the ADDV analysis dataset.\nThe code has access to libname _in (input data) and libname _out (output location).\nRead source datasets from _in librar...
CA056-015
ADDV
train
60
distilled
[ { "role": "system", "content": "You are a clinical SAS programming assistant." }, { "role": "user", "content": "Write SAS code to derive and build the ADCMB analysis dataset.\nThe code has access to libname _in (input data) and libname _out (output location).\nRead source datasets from _in libra...
CA056-015
ADCMB
train
110
distilled
[ { "role": "system", "content": "You are a clinical SAS programming assistant." }, { "role": "user", "content": "Write SAS code to derive and build the ADFER analysis dataset.\nThe code has access to libname _in (input data) and libname _out (output location).\nRead source datasets from _in libra...
CA056-015
ADFER
train
27
distilled
[ { "role": "system", "content": "You are a clinical SAS programming assistant." }, { "role": "user", "content": "Write SAS code to derive and build the ADGF analysis dataset.\nThe code has access to libname _in (input data) and libname _out (output location).\nRead source datasets from _in librar...
CA056-015
ADGF
train
111
distilled
[ { "role": "system", "content": "You are a clinical SAS programming assistant." }, { "role": "user", "content": "Write SAS code to derive and build the ADHRU analysis dataset.\nThe code has access to libname _in (input data) and libname _out (output location).\nRead source datasets from _in libra...
CA056-015
ADHRU
train
110
distilled
[ { "role": "system", "content": "You are a clinical SAS programming assistant." }, { "role": "user", "content": "Write SAS code to derive and build the ADHC analysis dataset.\nThe code has access to libname _in (input data) and libname _out (output location).\nRead source datasets from _in librar...
CA056-015
ADHC
train
110
distilled
[ { "role": "system", "content": "You are a clinical SAS programming assistant." }, { "role": "user", "content": "Write SAS code to derive and build the ADMH analysis dataset.\nThe code has access to libname _in (input data) and libname _out (output location).\nRead source datasets from _in librar...
CA056-015
ADMH
train
114
distilled
[{"role":"system","content":"You are a clinical SAS programming assistant."},{"role":"user","content(...TRUNCATED)
CA056-015
ADPC
train
51
distilled
[{"role":"system","content":"You are a clinical SAS programming assistant."},{"role":"user","content(...TRUNCATED)
CA057-008
*ALL*
train
13
distilled
[{"role":"system","content":"You are a clinical SAS programming assistant."},{"role":"user","content(...TRUNCATED)
CA057-003
ADEFTTES
train
60
distilled
End of preview. Expand in Data Studio

SASBuddyLM v2 — Clinical SAS Fine-tuning Dataset

ChatML-formatted spec-to-program pairs for fine-tuning a clinical SAS code assistant.

Splits (study-level partition)

  • train: 102
  • validation: 0
  • test: 100

No study appears in more than one split.

Record schema

Each JSONL row:

{
  "messages": [
    {"role": "system", ...},
    {"role": "user", ...},
    {"role": "assistant", ...}
  ],
  "study": "...",
  "dataset": "...",
  "split": "train|val|test"
}

Provenance

Records were distilled from per-variable specification-code fragments aggregated by (study, dataset). Distillation used GPT-5.4 with a prompt that rewrites fragments into self-contained SAS programs using _in/_out librefs.

Studies

  • Test: CA057-001, DR202403, CA116-001, IA112025, CA073-1020, BR335-2025, CA239-0012
  • Val: CA245-0001, IM011-1069, IM027-1015, CA239-0007
  • Train: CA056-015, CA057-003, CA057-008, CA099-003, CA224-1093, CA224-127, CA239-0010, CV027-031, CV027031, IM011-246, IM011-247, IM027-068
Downloads last month
4