Spaces:
Paused
Paused
File size: 2,207 Bytes
4ecb2ec | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 | [client]
name = "sqlbot"
task = "sql"
description = "A BI startup wants a small text to SQL model that runs on a CPU box next to their warehouse, answers with one SQL statement and nothing else, and is judged by whether the query returns the right rows."
[data]
min_chars = 40
max_chars = 6000
val_fraction = 0.0
max_records = 0
seed = 1
dedup = "exact"
num_perm = 64
bands = 8
shingle = 5
workers = 4
[[data.sources]]
name = "gretel_train"
url = "https://huggingface.co/datasets/gretelai/synthetic_text_to_sql/resolve/main/synthetic_text_to_sql_train.snappy.parquet"
format = "parquet"
split = "train"
[[data.sources]]
name = "gretel_test"
url = "https://huggingface.co/datasets/gretelai/synthetic_text_to_sql/resolve/main/synthetic_text_to_sql_test.snappy.parquet"
format = "parquet"
split = "val"
[tokenizer]
vocab_size = 8192
sample_bytes = 100000000
[model]
dim = 384
n_layers = 6
n_heads = 6
n_kv_heads = 2
seq_len = 768
rope_theta = 10000.0
[pretrain]
tokens = 80000000
batch_tokens = 98304
micro_batch = 32
lr = 0.001
min_lr = 0.0001
warmup_steps = 200
weight_decay = 0.1
grad_clip = 1.0
eval_every = 200
eval_batches = 20
checkpoint_every = 200
log_every = 20
dtype = "auto"
compile = false
seed = 1
[sft]
epochs = 2
lr = 0.0001
batch_size = 32
max_examples = 0
warmup_steps = 100
[dpo]
beta = 0.1
lr = 0.00001
pairs = 4000
epochs = 1
batch_size = 8
candidates = 4
temperature = 0.8
warmup_steps = 20
[generation]
max_new_tokens = 160
temperature = 0.0
top_p = 1.0
top_k = 0
repetition_penalty = 1.0
[eval]
samples = 1000
batch_size = 64
perplexity_batches = 50
[[eval.gates]]
metric = "execution_accuracy"
min = 0.4
[[eval.gates]]
metric = "valid_rate"
min = 0.8
[task]
timeout_ms = 2000
max_rows = 1000
[export]
quantize = "int8"
[serve]
system_prompt = "You translate questions about a database into one SQL statement. Reply with the SQL only."
stop = [";"]
example = "Schema:\nCREATE TABLE orders (id INT, customer TEXT, amount DECIMAL(10,2), placed_on DATE);\nINSERT INTO orders VALUES (1, 'Ana', 120.50, '2024-01-05'), (2, 'Ben', 80.00, '2024-02-11'), (3, 'Ana', 45.25, '2024-02-20');\n\nQuestion: What is the total order amount per customer?"
max_batch = 8
batch_wait_ms = 20
|