# sample = first 2000 of 20000 records; tokens per record = tokens / 2000 (prefix is cut at a line end, so per-record is approximate to <1 record)
view	src_bytes	bytes_per_record	cl100k_base:B/tok	cl100k_base:tok/rec	o200k_base:B/tok	o200k_base:tok/rec	gpt2:B/tok	gpt2:tok/rec	mistral-7b:B/tok	mistral-7b:tok/rec	qwen2.5:B/tok	qwen2.5:tok/rec
recs.json	230653	115.3	3.184	36.2	3.092	37.3	2.889	39.9	2.228	51.8	2.407	47.9
recs_pretty.json	368557	184.3	2.811	65.6	2.807	65.6	1.772	104.0	2.065	89.2	2.383	77.3
recs.jsonl	232602	116.3	3.098	37.5	3.011	38.6	2.751	42.3	2.186	53.2	2.358	49.3
recs.yaml	244551	122.3	2.377	51.4	2.365	51.7	2.264	54.0	1.804	67.8	1.934	63.2
recs.xml	288580	144.3	2.762	52.2	2.741	52.6	2.518	57.3	2.180	66.2	2.254	64.0
recs.csv	120581	60.3	2.631	22.9	2.602	23.2	2.124	28.4	1.613	37.4	1.732	34.8
recs.toml	250558	125.3	2.690	46.6	2.685	46.7	2.418	51.8	1.981	63.2	2.147	58.4
recs_table.md	152629	76.3	2.686	28.4	2.662	28.7	2.463	31.0	1.702	44.8	1.896	40.3
recs_list.md	192629	96.3	2.975	32.4	2.952	32.6	2.766	34.8	2.017	47.7	2.180	44.2
recs.bin as base64	61336	30.7	1.514	20.3	1.579	19.4	1.410	21.8	1.343	22.8	1.488	20.6
recs.bin as hex	92000	46.0	2.023	22.7	2.018	22.8	1.916	24.0	1.041	44.2	1.046	44.0
recs.msgpack as base64	221436	110.7	1.451	76.3	1.531	72.3	1.305	84.8	1.256	88.1	1.402	79.0