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": 76502.40131697868,
-      "required_pages_per_second_upper_estimate": 0.005292613633135749
+      "required_token_positions_per_second": 76420.14315721064,
+      "required_pages_per_second_upper_estimate": 0.00528692282277777
     },
     {
       "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": 9562.800164622335,
-      "required_pages_per_second_upper_estimate": 0.004003744734951975
+      "required_token_positions_per_second": 9552.51789465133,
+      "required_pages_per_second_upper_estimate": 0.003999439763233329
     },
     {
       "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": 19125.60032924467,
-      "required_pages_per_second_upper_estimate": 0.003983358172583166
+      "required_token_positions_per_second": 19105.03578930266,
+      "required_pages_per_second_upper_estimate": 0.003979075121236633
     },
     {
       "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": 28688.400493867004,
-      "required_pages_per_second_upper_estimate": 0.0017198258049561444
+      "required_token_positions_per_second": 28657.55368395399,
+      "required_pages_per_second_upper_estimate": 0.0017179765858022105
     },
     {
       "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": 28688.400493867004,
-      "required_pages_per_second_upper_estimate": 0.002230396951460234
+      "required_token_positions_per_second": 28657.55368395399,
+      "required_pages_per_second_upper_estimate": 0.0022279987476702743
     },
     {
       "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": 28688.400493867004,
-      "required_pages_per_second_upper_estimate": 0.36798399833079365
+      "required_token_positions_per_second": 28657.55368395399,
+      "required_pages_per_second_upper_estimate": 0.3675883285739535
     }
   ],
   "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": 0.04824891578905266,
-    "producer_p95_seconds": 0.16322065300028044,
-    "gpu_update_seconds": 0.9637344544848563,
-    "producer_batches_per_second": 20.725854325350312,
-    "gpu_batches_per_second": 1.037630226196,
-    "producer_headroom": 19.97421991197408,
+    "producer_batch_seconds": 0.07460991712502363,
+    "producer_p95_seconds": 0.1716337019997809,
+    "gpu_update_seconds": 0.9647718121690456,
+    "producer_batches_per_second": 13.403043972348915,
+    "gpu_batches_per_second": 1.0365145284994932,
+    "producer_headroom": 12.930879021784467,
     "sustainable": true,
-    "buffer_seconds": 60.715270632545945,
+    "buffer_seconds": 60.78062416664987,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 347300.0185245126,
-    "upload_bytes_per_second": 5171426.317954449,
+    "generation_bytes_per_second": 360394.6505101085,
+    "upload_bytes_per_second": 5085743.629061496,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.52249924470456,
+    "estimated_full_shards_per_hour": 77.33230244138186,
     "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": [
@@ -352,6 +352,18 @@
       "bytes": 7144629,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.0513503590000255,
+      "bytes": 1139099,
+      "pages": 34,
+      "shards": 2
+    },
+    {
+      "seconds": 0.996597190999637,
+      "bytes": 6046442,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -48,3 +48,4 @@
 {"step": 889800, "loss": 0.5502925515174866, "lr": 3e-05, "grad_norm": 0.3064795285302192, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130241963479, "tokens_per_second": 153270.26601528405, "data_wait_s": 1.0538000424276106e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 889900, "loss": 0.48231640458106995, "lr": 3e-05, "grad_norm": 0.3191409479241946, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130256694665, "tokens_per_second": 153171.3023264892, "data_wait_s": 8.26599989522947e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 890000, "loss": 0.5526643991470337, "lr": 3e-05, "grad_norm": 0.32601227166011965, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130271344653, "tokens_per_second": 152680.7386107045, "data_wait_s": 1.1977999747614376e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 890100, "loss": 0.4908837080001831, "lr": 3e-05, "grad_norm": 0.3075500939918485, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130285879941, "tokens_per_second": 139867.4475602635, "data_wait_s": 9.98499945126241e-06, "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 sha256:434158f4a5a26bf3ecb974d7feb03e1b3a5793737b6f0a88e6c3afd37f4e69e0
-size 106848
+oid sha256:60b56c036868d4d7a0f88b9c7852e378900cf343f0cbb22023a18a5196367ea8
+size 108966