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": 77322.00500718379,
-      "required_pages_per_second_upper_estimate": 0.00534931571816671
+      "required_token_positions_per_second": 76936.16085357314,
+      "required_pages_per_second_upper_estimate": 0.005322622124338142
     },
     {
       "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": 9665.250625897974,
-      "required_pages_per_second_upper_estimate": 0.0040466386036791735
+      "required_token_positions_per_second": 9617.020106696642,
+      "required_pages_per_second_upper_estimate": 0.004026445492457346
     },
     {
       "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": 19330.501251795948,
-      "required_pages_per_second_upper_estimate": 0.004026033631149893
+      "required_token_positions_per_second": 19234.040213393284,
+      "required_pages_per_second_upper_estimate": 0.004005943340699267
     },
     {
       "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": 28995.75187769392,
-      "required_pages_per_second_upper_estimate": 0.0017382510511181782
+      "required_token_positions_per_second": 28851.060320089928,
+      "required_pages_per_second_upper_estimate": 0.001729577013171035
     },
     {
       "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": 28995.75187769392,
-      "required_pages_per_second_upper_estimate": 0.002254292169656912
+      "required_token_positions_per_second": 28851.060320089928,
+      "required_pages_per_second_upper_estimate": 0.002243043037484103
     },
     {
       "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": 28995.75187769392,
-      "required_pages_per_second_upper_estimate": 0.37192637187432076
+      "required_token_positions_per_second": 28851.060320089928,
+      "required_pages_per_second_upper_estimate": 0.37007042393106715
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.4460236786093148,
-    "producer_p95_seconds": 0.19942733399966528,
-    "gpu_update_seconds": 0.9535189884580737,
-    "producer_batches_per_second": 2.2420334344534414,
-    "gpu_batches_per_second": 1.0487468127059434,
-    "producer_headroom": 2.1378214525092263,
+    "producer_batch_seconds": 0.3676429310781444,
+    "producer_p95_seconds": 0.19168011700003262,
+    "gpu_update_seconds": 0.9583010015319195,
+    "producer_batches_per_second": 2.7200305390543327,
+    "gpu_batches_per_second": 1.0435134664384378,
+    "producer_headroom": 2.606607989773174,
     "sustainable": true,
-    "buffer_seconds": 60.071696272858645,
+    "buffer_seconds": 60.37296309651093,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 313464.15864199266,
+    "generation_bytes_per_second": 284107.04530963453,
     "upload_bytes_per_second": 5289093.047335207,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.2621113724216,
+    "estimated_full_shards_per_hour": 60.96275824991967,
     "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": [

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -59,3 +59,4 @@
 {"step": 890900, "loss": 0.5388245582580566, "lr": 3e-05, "grad_norm": 0.3211956049093675, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130402871994, "tokens_per_second": 151829.41128575586, "data_wait_s": 1.174099998024758e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 891000, "loss": 0.5289643406867981, "lr": 3e-05, "grad_norm": 0.3304750209118958, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130417428723, "tokens_per_second": 137786.4988271708, "data_wait_s": 1.584900019224733e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 891100, "loss": 0.5342656970024109, "lr": 3e-05, "grad_norm": 0.3095160137590757, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130432070713, "tokens_per_second": 152782.13955704786, "data_wait_s": 1.8817000636772718e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 891200, "loss": 0.4989605247974396, "lr": 3e-05, "grad_norm": 0.31334013940086874, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130446723013, "tokens_per_second": 153155.12638318937, "data_wait_s": 1.2485999832279049e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:89405cf7b635d437fa7ad83118b1c1e9183989158f7eeaf75d8a8b6acde3b1ea
-size 130146
+oid sha256:22ca0ff34ee3dada1a1fb56c54b946fcc9e9f6986bbc8c85915b0a64e523bffb
+size 132264