Hugging Face's logo

Datasets:
piotrwilam
/
AtlasOpenCoder

DOI:
License:
Dataset card Files
xet
 Community
AtlasOpenCoder / validation /schedule /CHECKPOINTS.json
piotrwilam's picture
piotrwilam
AtlasOpenCoder v1: the battery, the artefacts and both papers' manifests
9b7bd53
Raw History Blame Contribute Delete
7.82 kB
{
"repo": "OpenCoder-LLM/OpenCoder-1.5B-Base-Checkpoints",
"tokens_per_step": 4194304,
"pretrain_steps": [
1000,
2000,
3000,
4000,
5000,
6000,
7000,
8000,
9000,
10000,
11000,
12000,
13000,
14000,
15000,
16000,
17000,
18000,
19000,
20000,
21000,
22000,
23000,
24000,
25000,
26000,
27000,
28000,
29000,
30000,
31000,
32000,
33000,
34000,
35000,
36000,
37000,
38000,
39000,
40000,
41000,
42000,
43000,
44000,
45000,
46000,
47000,
48000,
49000,
50000,
51000,
52000,
53000,
54000,
55000,
56000,
57000,
58000,
59000,
60000,
61000,
62000,
63000,
64000,
65000,
66000,
67000,
68000,
69000,
70000,
71000,
72000,
73000,
74000,
75000,
76000,
77000,
78000,
79000,
80000,
81000,
82000,
83000,
84000,
85000,
86000,
87000,
88000,
89000,
90000,
91000,
92000,
93000,
94000,
95000,
96000,
97000,
98000,
99000,
100000,
101000,
102000,
103000,
104000,
105000,
106000,
107000,
108000,
109000,
110000,
111000,
112000,
113000,
114000,
115000,
116000,
117000,
118000,
119000,
120000,
121000,
122000,
123000,
124000,
125000,
126000,
127000,
128000,
129000,
130000,
131000,
132000,
133000,
134000,
135000,
136000,
137000,
138000,
139000,
140000,
141000,
142000,
143000,
144000,
145000,
146000,
147000,
148000,
149000,
150000,
151000,
152000,
153000,
154000,
155000,
156000,
157000,
158000,
159000,
160000,
161000,
162000,
163000,
164000,
165000,
166000,
167000,
168000,
169000,
170000,
171000,
172000,
173000,
174000,
175000,
176000,
177000,
178000,
179000,
180000,
181000,
182000,
183000,
184000,
185000,
186000,
187000,
188000,
189000,
190000,
191000,
192000,
193000,
194000,
195000,
196000,
197000,
198000,
199000,
200000,
201000,
202000,
203000,
204000,
205000,
206000,
207000,
208000,
209000,
210000,
211000,
212000,
213000,
214000,
215000,
216000,
217000,
218000,
219000,
220000,
221000,
222000,
223000,
224000,
225000,
226000,
227000,
228000,
229000,
230000,
231000,
232000,
233000,
234000,
235000,
236000,
237000,
238000,
239000,
240000,
241000,
242000,
243000,
244000,
245000,
246000,
247000,
248000,
249000,
250000,
251000,
252000,
253000,
254000,
255000,
256000,
257000,
258000,
259000,
260000,
261000,
262000,
263000,
264000,
265000,
266000,
267000,
268000,
269000,
270000,
271000,
272000,
273000,
274000,
275000,
276000,
277000,
278000,
279000,
280000,
281000,
282000,
283000,
284000,
285000,
286000,
287000,
288000,
289000,
290000,
291000,
292000,
293000,
294000,
295000,
296000,
297000,
298000,
299000,
300000,
301000,
302000,
303000,
304000,
305000,
306000,
307000,
308000,
309000,
310000,
311000,
312000,
313000,
314000,
315000,
316000,
317000,
318000,
319000,
320000,
321000,
322000,
323000,
324000,
325000,
326000,
327000,
328000,
329000,
330000,
331000,
332000,
333000,
334000,
335000,
336000,
337000,
338000,
339000,
340000,
341000,
342000,
343000,
344000,
345000,
346000,
347000,
348000,
349000,
350000,
351000,
352000,
353000,
354000,
355000,
356000,
357000,
358000,
359000,
360000,
361000,
362000,
363000,
364000,
365000,
366000,
367000,
368000,
369000,
370000,
371000,
372000,
373000,
374000,
375000,
376000,
377000,
378000,
379000,
380000,
381000,
382000,
383000,
384000,
385000,
386000,
387000,
388000,
389000,
390000,
391000,
392000,
393000,
394000,
395000,
396000,
397000,
398000,
399000,
400000,
401000,
402000,
403000,
404000,
405000,
406000,
407000,
408000,
409000,
410000,
411000,
412000,
413000,
414000,
415000,
416000,
417000,
418000,
419000,
420000,
421000,
422000,
423000,
424000,
425000,
426000,
427000,
428000,
429000,
430000,
431000,
432000,
433000,
434000,
435000,
436000,
437000,
438000,
439000,
440000,
441000,
442000,
443000,
444000,
445000,
446000,
447000,
448000,
449000,
450000,
451000,
452000,
453000,
454000,
455000,
456000,
457000,
458000,
459000,
460000,
461000,
462000,
463000,
464000,
465000,
466000,
467000,
468000,
469000,
470000,
471000,
472000,
473000,
474000,
475000,
476000,
477000,
478000,
479000,
480000,
481000,
482000,
483000,
484000,
484865
],
"anneal_steps": [
1000,
2000,
3000,
4000,
5000,
6000,
7000,
8000,
9000,
10000,
11000,
12000,
13000,
14000,
15000,
16000,
17000,
18000,
19000,
20000,
21000,
22000,
23000,
23841
],
"fork_step": 373000,
"final_base_anneal_step": 23000,
"last_pretrain_tokens": 2033671208960,
"learning_rate_values": {
"standing": "measured from the tech report, 1 Oct 2026 (item 6's owed literature item)",
"source": "OpenCoder: The Open Cookbook for Top-Tier Code Large Language Models, arXiv:2411.04905v3, section 3 training details, the 1.5B paragraph",
"schedule": "WSD (warmup-stable-decay), after MiniCPM (Hu et al. 2024)",
"warmup_steps": 2000,
"warmup_tokens_reported": 8000000000.0,
"warmup_tokens_computed": 8388608000,
"peak_lr": 0.0003,
"stable_lr": 0.0003,
"stable_note": "constant at the peak after warm-up, for the whole pretraining branch",
"anneal_final_lr": 1e-05,
"anneal_decay": "exponential, during the annealing phase only",
"global_batch": 1024,
"micro_batch": 4,
"sequence_length": 4096,
"tokens_per_step_from_report": 4194304,
"epochs": 4,
"pretrain_tokens_reported": 2000000000000.0,
"anneal_tokens_reported": 100000000000.0
},
"learning_rate_note": "Measured 1 Oct. The schedule is WSD: 2,000 steps of warm-up to 3e-4, then CONSTANT at 3e-4 for 99.6% of the pretraining branch, and the exponential decay to 1e-5 happens only on the annealing branch. So there is no learning-rate drop inside the pretraining branch for C8's bend to align with, and C8's first clause is vacuous as written -- see deviation A11. Two further facts the schedule forces: the first reference, step 1,000 = 4.19B, sits INSIDE the warm-up, half way through it, at an LR of about 1.5e-4 and still rising -- it is not a constant-LR point; and the 2.0337T branch is four epochs over about 508B unique tokens, so the fork at 1.5645T is 8% into the fourth epoch while every dense reference at or below 84B is in the first. Tokens per step from the report, 1024 x 4096 = 4,194,304, equals the measured figure exactly.",
"reference_set": {
"dense": [
1000,
2000,
3000,
4000,
5000,
6000,
7000,
8000,
9000,
10000,
11000,
12000,
13000,
14000,
15000,
16000,
17000,
18000,
19000,
20000
],
"tail": [
25000,
31000,
39000,
49000,
61000,
76000,
95000,
119000,
149000,
186000,
233000,
291000,
364000,
455000,
484865
],
"anneal": [
1000,
2000,
3000,
4000,
5000,
6000,
7000,
8000,
9000,
10000,
11000,
12000,
13000,
14000,
15000,
16000,
17000,
18000,
19000,
20000,
21000,
22000,
23000,
23841
],
"n": 59
},
"epochs": {
"n": 4,
"tokens_per_epoch": 508417802240.0,
"boundaries_tokens": [
508417802240.0,
1016835604480.0,
1525253406720.0
],
"fork_epoch": 4,
"dense_references_all_in_epoch": 1
}
}