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": 76490.70180550122,
-      "required_pages_per_second_upper_estimate": 0.005291804233785132
+      "required_token_positions_per_second": 76621.02201348281,
+      "required_pages_per_second_upper_estimate": 0.005300820088157842
     },
     {
       "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": 9561.337725687652,
-      "required_pages_per_second_upper_estimate": 0.004003132442309221
+      "required_token_positions_per_second": 9577.627751685352,
+      "required_pages_per_second_upper_estimate": 0.004009952736020033
     },
     {
       "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": 19122.675451375304,
-      "required_pages_per_second_upper_estimate": 0.003982748997657193
+      "required_token_positions_per_second": 19155.255503370703,
+      "required_pages_per_second_upper_estimate": 0.003989534563294085
     },
     {
       "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": 28684.01317706296,
-      "required_pages_per_second_upper_estimate": 0.0017195627920127867
+      "required_token_positions_per_second": 28732.88325505606,
+      "required_pages_per_second_upper_estimate": 0.0017224924785681846
     },
     {
       "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": 28684.01317706296,
-      "required_pages_per_second_upper_estimate": 0.0022300558568764863
+      "required_token_positions_per_second": 28732.88325505606,
+      "required_pages_per_second_upper_estimate": 0.0022338552904836943
     },
     {
       "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": 28684.01317706296,
-      "required_pages_per_second_upper_estimate": 0.36792772254156514
+      "required_token_positions_per_second": 28732.88325505606,
+      "required_pages_per_second_upper_estimate": 0.3685545754294591
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.0836121916953374,
-    "producer_p95_seconds": 0.241298930000994,
-    "gpu_update_seconds": 0.9638818609283236,
-    "producer_batches_per_second": 11.959978320431524,
-    "gpu_batches_per_second": 1.037471541415761,
-    "producer_headroom": 11.528006160159944,
+    "producer_batch_seconds": 0.12949761582031272,
+    "producer_p95_seconds": 0.2468350390008709,
+    "gpu_update_seconds": 0.962242450734033,
+    "producer_batches_per_second": 7.722149891837176,
+    "gpu_batches_per_second": 1.039239122361692,
+    "producer_headroom": 7.430580436856952,
     "sustainable": true,
-    "buffer_seconds": 60.72455723848439,
+    "buffer_seconds": 60.62127439624408,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 413992.4195273213,
-    "upload_bytes_per_second": 5504252.0436081635,
+    "generation_bytes_per_second": 461182.7129237697,
+    "upload_bytes_per_second": 5507323.355854074,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 88.83313598026972,
+    "estimated_full_shards_per_hour": 98.95907440934009,
     "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": 2.0578567419997853,
-      "bytes": 11618530,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 1.2497664069996972,
       "bytes": 1138477,
@@ -562,6 +556,12 @@
       "bytes": 2658914,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 0.9456378320010117,
+      "bytes": 5728053,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -144,3 +144,4 @@
 {"step": 411500, "loss": 0.6054062843322754, "lr": 0.0003, "grad_norm": 0.17300995907106745, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60233332164, "tokens_per_second": 153308.91245792247, "data_wait_s": 1.547600004414562e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 411600, "loss": 0.6161276698112488, "lr": 0.0003, "grad_norm": 0.16131318474121076, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60248041387, "tokens_per_second": 152322.1890086199, "data_wait_s": 1.2078000509063713e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 411700, "loss": 0.5794970393180847, "lr": 0.0003, "grad_norm": 0.1733984960715895, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60262659478, "tokens_per_second": 151426.0750094452, "data_wait_s": 1.756699930410832e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 411800, "loss": 0.6317907571792603, "lr": 0.0003, "grad_norm": 0.16715218412259703, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60277173159, "tokens_per_second": 150989.12647197262, "data_wait_s": 1.9148999854223803e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:25e8bc2a0a0242e8939aa819f0cbd3cd00c9d20269076411e6da0ac6e047e126
-size 311742
+oid sha256:f88d03ce5f57d6cbe7e1e6dd683e9710b928ac471a6fd967a2c53cce2f6661d9
+size 313860