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": 76783.62942529761,
-      "required_pages_per_second_upper_estimate": 0.005312069646208366
+      "required_token_positions_per_second": 76418.14584616253,
+      "required_pages_per_second_upper_estimate": 0.005286784644165062
     },
     {
       "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": 9597.953678162201,
-      "required_pages_per_second_upper_estimate": 0.004018462776982278
+      "required_token_positions_per_second": 9552.268230770316,
+      "required_pages_per_second_upper_estimate": 0.003999335234179923
     },
     {
       "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": 19195.907356324402,
-      "required_pages_per_second_upper_estimate": 0.003998001272202885
+      "required_token_positions_per_second": 19104.536461540632,
+      "required_pages_per_second_upper_estimate": 0.003978971124431962
     },
     {
       "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": 28793.8610344866,
-      "required_pages_per_second_upper_estimate": 0.0017261480033373673
+      "required_token_positions_per_second": 28656.80469231095,
+      "required_pages_per_second_upper_estimate": 0.0017179316848968545
     },
     {
       "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": 28793.8610344866,
-      "required_pages_per_second_upper_estimate": 0.0022385960446215125
+      "required_token_positions_per_second": 28656.80469231095,
+      "required_pages_per_second_upper_estimate": 0.002227940516864494
     },
     {
       "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": 28793.8610344866,
-      "required_pages_per_second_upper_estimate": 0.36933673291115554
+      "required_token_positions_per_second": 28656.80469231095,
+      "required_pages_per_second_upper_estimate": 0.36757872131336117
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.06402123775765745,
-    "producer_p95_seconds": 0.2285809569984849,
-    "gpu_update_seconds": 0.9602046758121742,
-    "producer_batches_per_second": 15.619816720591162,
-    "gpu_batches_per_second": 1.041444626536697,
-    "producer_headroom": 14.998221050440815,
+    "producer_batch_seconds": 0.12741619256237868,
+    "producer_p95_seconds": 0.22349988399946596,
+    "gpu_update_seconds": 0.9647970280307759,
+    "producer_batches_per_second": 7.848296043773507,
+    "gpu_batches_per_second": 1.0364874382346263,
+    "producer_headroom": 7.572012698138376,
     "sustainable": true,
-    "buffer_seconds": 60.492894576166975,
+    "buffer_seconds": 60.78221276593889,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 363769.6830396918,
-    "upload_bytes_per_second": 5191802.552146504,
+    "generation_bytes_per_second": 360152.3332979293,
+    "upload_bytes_per_second": 5227697.790039282,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.05650585549418,
+    "estimated_full_shards_per_hour": 77.28030680850419,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
-    {
-      "seconds": 0.8282303630003298,
-      "bytes": 647721,
-      "pages": 20,
-      "shards": 1
-    },
     {
       "seconds": 1.3377682740010641,
       "bytes": 13564102,
@@ -562,6 +556,12 @@
       "bytes": 5054764,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.5159087600004568,
+      "bytes": 7203777,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -195,3 +195,4 @@
 {"step": 623900, "loss": 0.55637127161026, "lr": 0.0001, "grad_norm": 0.17060754407015657, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91322495048, "tokens_per_second": 152848.3170181749, "data_wait_s": 1.3158998626749963e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 624000, "loss": 0.5446088910102844, "lr": 0.0001, "grad_norm": 0.15555424520469768, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91337146397, "tokens_per_second": 151456.037639442, "data_wait_s": 1.241100108018145e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 624100, "loss": 0.5719287991523743, "lr": 0.0001, "grad_norm": 0.17895209546245056, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91351703776, "tokens_per_second": 151328.9869741156, "data_wait_s": 1.020499985315837e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 624200, "loss": 0.5563517808914185, "lr": 0.0001, "grad_norm": 0.17141695445517058, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91366226013, "tokens_per_second": 150570.36739884503, "data_wait_s": 6.350001058308408e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9b6f274ff87962225b78ddb089800876456cf71d9d7f0939ce43e9950bea505e
-size 420543
+oid sha256:5e1d54ae5e77684e83951888953ab93b58aba65f25dff260ddf9e837baa999d4
+size 422661