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": 73906.08550783987,
-      "required_pages_per_second_upper_estimate": 0.006405992178558342
+      "required_token_positions_per_second": 74571.31616966445,
+      "required_pages_per_second_upper_estimate": 0.006463652686313598
     },
     {
       "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": 9238.260688479984,
-      "required_pages_per_second_upper_estimate": 0.003685280312940795
+      "required_token_positions_per_second": 9321.414521208057,
+      "required_pages_per_second_upper_estimate": 0.0037184516200766144
     },
     {
       "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": 18476.52137695997,
-      "required_pages_per_second_upper_estimate": 0.004274218044889582
+      "required_token_positions_per_second": 18642.829042416113,
+      "required_pages_per_second_upper_estimate": 0.004312690396377916
     },
     {
       "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": 27714.782065439955,
-      "required_pages_per_second_upper_estimate": 0.0016876848195615354
+      "required_token_positions_per_second": 27964.243563624168,
+      "required_pages_per_second_upper_estimate": 0.0017028757159776244
     },
     {
       "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": 27714.782065439955,
-      "required_pages_per_second_upper_estimate": 0.0021335606428245866
+      "required_token_positions_per_second": 27964.243563624168,
+      "required_pages_per_second_upper_estimate": 0.002152764879508432
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27714.782065439955,
-      "required_pages_per_second_upper_estimate": 0.3441248378439718
+      "required_token_positions_per_second": 27964.243563624168,
+      "required_pages_per_second_upper_estimate": 0.34722231475749243
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.3866974157733978,
-    "producer_p95_seconds": 0.33053609099806636,
-    "gpu_update_seconds": 0.9975903810002089,
-    "producer_batches_per_second": 2.5860012485472454,
-    "gpu_batches_per_second": 1.0024154392881928,
-    "producer_headroom": 2.5797699708052626,
+    "producer_batch_seconds": 0.08035000478147936,
+    "producer_p95_seconds": 0.3311891699995613,
+    "gpu_update_seconds": 0.9886911454298897,
+    "producer_batches_per_second": 12.445549975007589,
+    "gpu_batches_per_second": 1.0114382075963604,
+    "producer_headroom": 12.304805060295188,
     "sustainable": true,
-    "buffer_seconds": 60.85301324101275,
+    "buffer_seconds": 60.31015987122327,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 362226.8170411782,
-    "upload_bytes_per_second": 6203472.035397348,
+    "generation_bytes_per_second": 360841.50211958133,
+    "upload_bytes_per_second": 6216050.0753486585,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.7254427282954,
+    "estimated_full_shards_per_hour": 77.42818639460162,
     "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": 1.3725059809985396,
-      "bytes": 7833557,
-      "pages": 41,
-      "shards": 3
-    },
     {
       "seconds": 1.04416848999972,
       "bytes": 7007035,
@@ -562,6 +556,12 @@
       "bytes": 7525026,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 0.7792704010025773,
+      "bytes": 5012728,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -278,3 +278,4 @@
 {"step": 27800, "loss": 0.6640053652226925, "lr": 0.0003, "grad_norm": 0.13030782590293608, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4068203688, "tokens_per_second": 149845.6373360764, "data_wait_s": 1.3565000699600205e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 27900, "loss": 0.6537720188498497, "lr": 0.0003, "grad_norm": 0.14147428053710845, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4082833473, "tokens_per_second": 148795.48222518666, "data_wait_s": 1.0709998605307192e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 28000, "loss": 0.7334572561085224, "lr": 0.0003, "grad_norm": 0.16861153780658747, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4097369174, "tokens_per_second": 139404.03716623955, "data_wait_s": 1.0496001777937636e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 28100, "loss": 0.6417201161384583, "lr": 0.0003, "grad_norm": 0.13232709154429506, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4112073984, "tokens_per_second": 150423.2972179369, "data_wait_s": 8.861999958753586e-06, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4efd49d795dc02a64a9b4bf90c2749afe64e6c21b16015c45c6a100d41a7b6de
-size 591346
+oid sha256:8d07344b06c5b0adbf77ee2d0d658ca99b1b7267fdc31aae4155f8d4bf3f8c7d
+size 593464