Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76119.66301634122,
-      "required_pages_per_second_upper_estimate": 0.0065978594667343345
+      "required_token_positions_per_second": 76197.96714523742,
+      "required_pages_per_second_upper_estimate": 0.006604646670167057
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9514.957877042652,
-      "required_pages_per_second_upper_estimate": 0.0037956589584500766
+      "required_token_positions_per_second": 9524.745893154677,
+      "required_pages_per_second_upper_estimate": 0.0037995635444210457
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19029.915754085305,
-      "required_pages_per_second_upper_estimate": 0.004402236094629134
+      "required_token_positions_per_second": 19049.491786309354,
+      "required_pages_per_second_upper_estimate": 0.004406764665157767
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28544.873631127954,
-      "required_pages_per_second_upper_estimate": 0.0017382330407580736
+      "required_token_positions_per_second": 28574.23767946403,
+      "required_pages_per_second_upper_estimate": 0.0017400211572404854
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28544.873631127954,
-      "required_pages_per_second_upper_estimate": 0.0021974633893917708
+      "required_token_positions_per_second": 28574.23767946403,
+      "required_pages_per_second_upper_estimate": 0.0021997239151175705
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28544.873631127954,
-      "required_pages_per_second_upper_estimate": 0.35443179695205873
+      "required_token_positions_per_second": 28574.23767946403,
+      "required_pages_per_second_upper_estimate": 0.35479640015724484
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,17 +155,17 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.049429777875019454,
-    "producer_p95_seconds": 0.1342726110001422,
-    "gpu_update_seconds": 0.9685802206477481,
-    "producer_batches_per_second": 20.230720083922822,
-    "gpu_batches_per_second": 1.032439005755496,
-    "producer_headroom": 19.595075322748794,
+    "producer_batch_seconds": 0.05407736080466563,
+    "producer_p95_seconds": 0.1416256800002884,
+    "gpu_update_seconds": 0.9675848682350079,
+    "producer_batches_per_second": 18.492026702488836,
+    "gpu_batches_per_second": 1.0335010734759849,
+    "producer_headroom": 17.89260522032591,
     "sustainable": true,
-    "buffer_seconds": 60.05197368016038,
+    "buffer_seconds": 60.9578466988055,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,

metrics/void-byte-v1/1790696019647648547-3fdea0a8/train.jsonl

@@ -18,3 +18,4 @@
 {"step": 276800, "loss": 0.6234837397933006, "lr": 0.0003, "grad_norm": 0.17697912234679625, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40515274660, "tokens_per_second": 153003.5590317808, "data_wait_s": 7.983000159583753e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 276900, "loss": 0.6361541375517845, "lr": 0.0003, "grad_norm": 0.14672465720243766, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40530044420, "tokens_per_second": 152390.50278827036, "data_wait_s": 1.304199986407184e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 277000, "loss": 0.6080015897750854, "lr": 0.0003, "grad_norm": 0.1453749490976164, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40544820615, "tokens_per_second": 152366.88545407142, "data_wait_s": 1.2525999864010373e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 277100, "loss": 0.561845388263464, "lr": 0.0003, "grad_norm": 0.13909139895340553, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40559593939, "tokens_per_second": 151182.7518431791, "data_wait_s": 1.3313000181369716e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790696019.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.22407.0.1790696019647648547-3fdea0a8

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4593a1a6c6268f8394d2d2912f85ffdf3740835a1965efbe5f3af961f01fef56
-size 42525
+oid sha256:b08001cd975307098ea8ec9c8b58816c12e68bcc145badd3ad3feabf6b0216b9
+size 44643