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": 77175.47242258527,
-      "required_pages_per_second_upper_estimate": 0.005339178253961744
+      "required_token_positions_per_second": 76561.19447506017,
+      "required_pages_per_second_upper_estimate": 0.005296681080230748
     },
     {
       "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": 9646.934052823159,
-      "required_pages_per_second_upper_estimate": 0.004038969837026289
+      "required_token_positions_per_second": 9570.149309382521,
+      "required_pages_per_second_upper_estimate": 0.004006821668395472
     },
     {
       "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": 19293.868105646317,
-      "required_pages_per_second_upper_estimate": 0.004018403912887955
+      "required_token_positions_per_second": 19140.298618765042,
+      "required_pages_per_second_upper_estimate": 0.003986419438670299
     },
     {
       "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": 28940.802158469476,
-      "required_pages_per_second_upper_estimate": 0.0017349568993540359
+      "required_token_positions_per_second": 28710.447928147565,
+      "required_pages_per_second_upper_estimate": 0.001721147514976781
     },
     {
       "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": 28940.802158469476,
-      "required_pages_per_second_upper_estimate": 0.002250020070685483
+      "required_token_positions_per_second": 28710.447928147565,
+      "required_pages_per_second_upper_estimate": 0.002232111042499133
     },
     {
       "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": 28940.802158469476,
-      "required_pages_per_second_upper_estimate": 0.3712215358765213
+      "required_token_positions_per_second": 28710.447928147565,
+      "required_pages_per_second_upper_estimate": 0.3682667991450541
     }
   ],
   "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": 0.14964189717233012,
-    "producer_p95_seconds": 0.37099440700694686,
-    "gpu_update_seconds": 0.9553294289705394,
-    "producer_batches_per_second": 6.682620435160503,
-    "gpu_batches_per_second": 1.046759337328902,
-    "producer_headroom": 6.384103964348741,
+    "producer_batch_seconds": 0.31144588858603584,
+    "producer_p95_seconds": 0.3554697510044207,
+    "gpu_update_seconds": 0.9629943799272478,
+    "producer_batches_per_second": 3.210830634303761,
+    "gpu_batches_per_second": 1.0384276594382076,
+    "producer_headroom": 3.0920118557327623,
     "sustainable": true,
-    "buffer_seconds": 60.185754025143986,
+    "buffer_seconds": 60.66864593541661,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 298859.1830600483,
+    "generation_bytes_per_second": 292173.39620120096,
     "upload_bytes_per_second": 4467479.035478239,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.12822359896744,
+    "estimated_full_shards_per_hour": 62.69360937621137,
     "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-v1/1791039222312944374-63c10d34/train.jsonl

@@ -423,3 +423,4 @@
 {"step": 646700, "loss": 0.5114997029304504, "lr": 0.0001, "grad_norm": 0.17141576251459942, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94660486978, "tokens_per_second": 152009.3076243558, "data_wait_s": 1.0433002898935229e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 646800, "loss": 0.5721449255943298, "lr": 0.0001, "grad_norm": 0.17860234885716397, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94675145228, "tokens_per_second": 151517.47726566327, "data_wait_s": 1.3228003808762878e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 646900, "loss": 0.557778000831604, "lr": 0.0001, "grad_norm": 0.1944920305595768, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94689630493, "tokens_per_second": 149477.76799090422, "data_wait_s": 1.7053003830369562e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 647000, "loss": 0.5787025690078735, "lr": 0.0001, "grad_norm": 0.1755085000784506, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94704061158, "tokens_per_second": 149091.65776771074, "data_wait_s": 9.91400156635791e-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 sha256:8b8c850ebb3ec20541b863ace42502ba49c8f89ba56438edd2be7f6631c766ad
-size 907362
+oid sha256:0bfe234d63bc59102629ee043eb688a0bf3e8bce6f93c147c6beaf561b53f42e
+size 909480