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": 76523.5510032188,
-      "required_pages_per_second_upper_estimate": 0.0052940768175038745
+      "required_token_positions_per_second": 76210.94592057366,
+      "required_pages_per_second_upper_estimate": 0.005272450072542759
     },
     {
       "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": 9565.44387540235,
-      "required_pages_per_second_upper_estimate": 0.004004851601448609
+      "required_token_positions_per_second": 9526.368240071708,
+      "required_pages_per_second_upper_estimate": 0.003988491448927724
     },
     {
       "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": 19130.8877508047,
-      "required_pages_per_second_upper_estimate": 0.003984459403055436
+      "required_token_positions_per_second": 19052.736480143416,
+      "required_pages_per_second_upper_estimate": 0.003968182554364294
     },
     {
       "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": 28696.331626207048,
-      "required_pages_per_second_upper_estimate": 0.0017203012642297922
+      "required_token_positions_per_second": 28579.10472021512,
+      "required_pages_per_second_upper_estimate": 0.0017132736902106974
     },
     {
       "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": 28696.331626207048,
-      "required_pages_per_second_upper_estimate": 0.002231013562114308
+      "required_token_positions_per_second": 28579.10472021512,
+      "required_pages_per_second_upper_estimate": 0.002221899685800102
     },
     {
       "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": 28696.331626207048,
-      "required_pages_per_second_upper_estimate": 0.3680857303806653
+      "required_token_positions_per_second": 28579.10472021512,
+      "required_pages_per_second_upper_estimate": 0.3665820694990459
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.09025825876565818,
-    "producer_p95_seconds": 0.2618322130001616,
-    "gpu_update_seconds": 0.963468096206079,
-    "producer_batches_per_second": 11.079318542986162,
-    "gpu_batches_per_second": 1.037917087174734,
-    "producer_headroom": 10.674569943871587,
+    "producer_batch_seconds": 0.10627905818753902,
+    "producer_p95_seconds": 0.2828239880000183,
+    "gpu_update_seconds": 0.9674200878813214,
+    "producer_batches_per_second": 9.409191397193316,
+    "gpu_batches_per_second": 1.0336771093827808,
+    "producer_headroom": 9.10264076836493,
     "sustainable": true,
-    "buffer_seconds": 60.69849006098298,
+    "buffer_seconds": 60.94746553652325,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 478369.98921515443,
-    "upload_bytes_per_second": 5205328.9185365,
+    "generation_bytes_per_second": 476352.04791976704,
+    "upload_bytes_per_second": 5222886.505184766,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 102.64706380215621,
+    "estimated_full_shards_per_hour": 102.21406057543524,
     "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": 0.9887657250001212,
-      "bytes": 1035853,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.3170902919991931,
       "bytes": 7309091,
@@ -562,6 +556,12 @@
       "bytes": 5942943,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 0.9916904240017175,
+      "bytes": 2484673,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -227,3 +227,4 @@
 {"step": 742700, "loss": 0.5572172999382019, "lr": 5e-05, "grad_norm": 0.23299492708085032, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108711199729, "tokens_per_second": 152683.98329650206, "data_wait_s": 1.0024999937741086e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 742800, "loss": 0.5505469441413879, "lr": 5e-05, "grad_norm": 0.22851434665660164, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108725881716, "tokens_per_second": 150772.91446298035, "data_wait_s": 1.5263001841958612e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 742900, "loss": 0.5111637115478516, "lr": 5e-05, "grad_norm": 0.22492677924003393, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108740399764, "tokens_per_second": 149454.1128898666, "data_wait_s": 1.6869002138264477e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 743000, "loss": 0.5352374315261841, "lr": 5e-05, "grad_norm": 0.22055568912012133, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108754935247, "tokens_per_second": 148846.01957060315, "data_wait_s": 1.2898999557364732e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:613caf27e71cd83804c8a729b3437317b5839789fce0651e632e2039542906a5
-size 488319
+oid sha256:9e09e6b1ee9ffb430915397232b82fc973016b8bcce84a91cede4ec9e1d84399
+size 490437