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": 75238.60688025369,
-      "required_pages_per_second_upper_estimate": 0.006521491753874629
+      "required_token_positions_per_second": 75060.10911632935,
+      "required_pages_per_second_upper_estimate": 0.0065060200200961135
     },
     {
       "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": 9404.825860031711,
-      "required_pages_per_second_upper_estimate": 0.003751725650244021
+      "required_token_positions_per_second": 9382.513639541168,
+      "required_pages_per_second_upper_estimate": 0.0037428249718929186
     },
     {
       "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": 18809.651720063423,
-      "required_pages_per_second_upper_estimate": 0.004351281886872771
+      "required_token_positions_per_second": 18765.027279082336,
+      "required_pages_per_second_upper_estimate": 0.004340958807815134
     },
     {
       "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": 28214.477580095132,
-      "required_pages_per_second_upper_estimate": 0.001718113654704286
+      "required_token_positions_per_second": 28147.540918623505,
+      "required_pages_per_second_upper_estimate": 0.0017140375632101854
     },
     {
       "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": 28214.477580095132,
-      "required_pages_per_second_upper_estimate": 0.0021720285867884584
+      "required_token_positions_per_second": 28147.540918623505,
+      "required_pages_per_second_upper_estimate": 0.002166875617295834
     },
     {
       "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": 28214.477580095132,
-      "required_pages_per_second_upper_estimate": 0.35032938376268213
+      "required_token_positions_per_second": 28147.540918623505,
+      "required_pages_per_second_upper_estimate": 0.34949825444979954
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.2749245478985358,
-    "producer_p95_seconds": 0.3486719409993384,
-    "gpu_update_seconds": 0.9799224501503876,
-    "producer_batches_per_second": 3.6373616239211275,
-    "gpu_batches_per_second": 1.0204889171041354,
-    "producer_headroom": 3.564332314595784,
+    "producer_batch_seconds": 0.09858272075774721,
+    "producer_p95_seconds": 0.34556575500027975,
+    "gpu_update_seconds": 0.9822527687208018,
+    "producer_batches_per_second": 10.143765482567229,
+    "gpu_batches_per_second": 1.018067886234936,
+    "producer_headroom": 9.963741730506161,
     "sustainable": true,
-    "buffer_seconds": 60.75519190932403,
+    "buffer_seconds": 60.89967166068971,
     "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": 382143.40680494235,
-    "upload_bytes_per_second": 4064991.715823895,
+    "generation_bytes_per_second": 382678.59378970717,
+    "upload_bytes_per_second": 4112634.810826324,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.99907925711825,
+    "estimated_full_shards_per_hour": 82.1139179255334,
     "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": 1.0883352769997146,
-      "bytes": 6129558,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.048354777998611,
       "bytes": 2655724,
@@ -562,6 +556,12 @@
       "bytes": 2586941,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 1.1623649149987614,
+      "bytes": 11298342,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -220,3 +220,4 @@
 {"step": 107000, "loss": 0.6474979519844055, "lr": 0.0003, "grad_norm": 0.13514128084320207, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15657524508, "tokens_per_second": 146448.4214527384, "data_wait_s": 1.5558998711640015e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 107100, "loss": 0.5815568156540394, "lr": 0.0003, "grad_norm": 0.14148197939184085, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15672075799, "tokens_per_second": 147734.25784706324, "data_wait_s": 1.8431001080898568e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 107200, "loss": 0.597231900319457, "lr": 0.0003, "grad_norm": 0.13108384581411894, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15686710105, "tokens_per_second": 148101.21162240565, "data_wait_s": 1.497999983257614e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 107300, "loss": 0.6110148895531893, "lr": 0.0003, "grad_norm": 0.12673179723575592, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15701488159, "tokens_per_second": 149519.10953555114, "data_wait_s": 2.0088002202101052e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d011b5cf9cc46676aac0e9cd51f1e128ecd40f22884495c51477e8c035fbfbfc
-size 473493
+oid sha256:1c905f3518c748ccdcdf756ac85e172806cdf31652f3b2301e41dd2010f7711b
+size 475611