Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76742.14030349927,
-      "required_pages_per_second_upper_estimate": 0.005309199332494332
+      "required_token_positions_per_second": 76859.79012147165,
+      "required_pages_per_second_upper_estimate": 0.005317338619886849
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9592.767537937409,
-      "required_pages_per_second_upper_estimate": 0.004016291448369081
+      "required_token_positions_per_second": 9607.473765183957,
+      "required_pages_per_second_upper_estimate": 0.0040224486386162646
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19185.535075874817,
-      "required_pages_per_second_upper_estimate": 0.003995840999720667
+      "required_token_positions_per_second": 19214.947530367914,
+      "required_pages_per_second_upper_estimate": 0.004001966838332993
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28778.302613812226,
-      "required_pages_per_second_upper_estimate": 0.0017252152997742716
+      "required_token_positions_per_second": 28822.42129555187,
+      "required_pages_per_second_upper_estimate": 0.0017278601473792368
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28778.302613812226,
-      "required_pages_per_second_upper_estimate": 0.00223738644584834
+      "required_token_positions_per_second": 28822.42129555187,
+      "required_pages_per_second_upper_estimate": 0.0022408164792960253
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28778.302613812226,
-      "required_pages_per_second_upper_estimate": 0.3691371661960753
+      "required_token_positions_per_second": 28822.42129555187,
+      "required_pages_per_second_upper_estimate": 0.3697030732744817
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,17 +158,17 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.04785211955469393,
-    "producer_p95_seconds": 0.13824637200013967,
-    "gpu_update_seconds": 0.9607237914973576,
-    "producer_batches_per_second": 20.897715906962528,
-    "gpu_batches_per_second": 1.0408818943074445,
-    "producer_headroom": 20.076932859771684,
+    "producer_batch_seconds": 0.05873504530470797,
+    "producer_p95_seconds": 0.15259553900023093,
+    "gpu_update_seconds": 0.9592532048744594,
+    "producer_batches_per_second": 17.02561043091328,
+    "gpu_batches_per_second": 1.0424776220902732,
+    "producer_headroom": 16.33187137079759,
     "sustainable": true,
-    "buffer_seconds": 60.52559886433353,
+    "buffer_seconds": 60.432951907090946,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -31,3 +31,4 @@
 {"step": 723100, "loss": 0.5263711214065552, "lr": 5e-05, "grad_norm": 0.22687436973325104, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105842314080, "tokens_per_second": 153348.85338377196, "data_wait_s": 1.1264000022492837e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 723200, "loss": 0.5374260544776917, "lr": 5e-05, "grad_norm": 0.21623103004222644, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105856894818, "tokens_per_second": 151238.93709425136, "data_wait_s": 1.2411000170686748e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 723300, "loss": 0.5372741222381592, "lr": 5e-05, "grad_norm": 0.20822957327548294, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105871498737, "tokens_per_second": 151807.38825043233, "data_wait_s": 1.5579999853798654e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 723400, "loss": 0.5299516320228577, "lr": 5e-05, "grad_norm": 0.23368338442515782, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105885999701, "tokens_per_second": 150844.57008442422, "data_wait_s": 1.045400040311506e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:30f03240c50bcd692048285062e845d2a1a13d278a9eb5294c1cfea3af036645
-size 70059
+oid sha256:8ff08aaa42e28f884f4266b4880fa4b6cfc3baf4aec5b4ed727ca14e6d5ab841
+size 72177