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": 76467.28831804452,
-      "required_pages_per_second_upper_estimate": 0.005290184434396102
+      "required_token_positions_per_second": 76214.30466202498,
+      "required_pages_per_second_upper_estimate": 0.005272682438069715
     },
     {
       "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": 9558.411039755565,
-      "required_pages_per_second_upper_estimate": 0.004001907100025608
+      "required_token_positions_per_second": 9526.788082753123,
+      "required_pages_per_second_upper_estimate": 0.003988667228291121
     },
     {
       "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": 19116.82207951113,
-      "required_pages_per_second_upper_estimate": 0.0039815298946616826
+      "required_token_positions_per_second": 19053.576165506245,
+      "required_pages_per_second_upper_estimate": 0.003968357438681379
     },
     {
       "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": 28675.233119266693,
-      "required_pages_per_second_upper_estimate": 0.0017190364409542726
+      "required_token_positions_per_second": 28580.364248259368,
+      "required_pages_per_second_upper_estimate": 0.001713349196993762
     },
     {
       "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": 28675.233119266693,
-      "required_pages_per_second_upper_estimate": 0.0022293732460022197
+      "required_token_positions_per_second": 28580.364248259368,
+      "required_pages_per_second_upper_estimate": 0.0022219976085654635
     },
     {
       "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": 28675.233119266693,
-      "required_pages_per_second_upper_estimate": 0.3678151013874462
+      "required_token_positions_per_second": 28580.364248259368,
+      "required_pages_per_second_upper_estimate": 0.3665982253724217
     }
   ],
   "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": 0.42068129538284893,
-    "producer_p95_seconds": 0.2977168880024692,
-    "gpu_update_seconds": 0.9641769915175858,
-    "producer_batches_per_second": 2.3770964171105615,
-    "gpu_batches_per_second": 1.0371539756679216,
-    "producer_headroom": 2.2919416719968937,
+    "producer_batch_seconds": 0.10846716339102613,
+    "producer_p95_seconds": 0.29848488599964185,
+    "gpu_update_seconds": 0.9673774539694275,
+    "producer_batches_per_second": 9.219380029281133,
+    "gpu_batches_per_second": 1.0337226652292886,
+    "producer_headroom": 8.918620379902569,
     "sustainable": true,
-    "buffer_seconds": 60.74315046560791,
+    "buffer_seconds": 60.944779600073936,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 421673.47086039034,
-    "upload_bytes_per_second": 4904213.867151546,
+    "generation_bytes_per_second": 422676.3614830114,
+    "upload_bytes_per_second": 4898378.845196422,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.48131079062254,
+    "estimated_full_shards_per_hour": 90.69650777213818,
     "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": 1.302292963002401,
-      "bytes": 11974442,
-      "pages": 40,
-      "shards": 4
-    },
     {
       "seconds": 1.1877205400014645,
       "bytes": 6194082,
@@ -562,6 +556,12 @@
       "bytes": 5386870,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.3874870750005357,
+      "bytes": 11872388,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -333,3 +333,4 @@
 {"step": 793300, "loss": 0.5124537944793701, "lr": 3e-05, "grad_norm": 0.27451279864426575, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116116564675, "tokens_per_second": 153188.23897549455, "data_wait_s": 8.362003427464515e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 793400, "loss": 0.5702718496322632, "lr": 3e-05, "grad_norm": 0.2712648949898607, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116131171122, "tokens_per_second": 151034.46029946272, "data_wait_s": 9.796996891964227e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 793500, "loss": 0.5302921533584595, "lr": 3e-05, "grad_norm": 0.24440704053070145, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116145981237, "tokens_per_second": 153521.0138528239, "data_wait_s": 1.4961005945224315e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 793600, "loss": 0.5755289793014526, "lr": 3e-05, "grad_norm": 0.3006565562961368, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116160663144, "tokens_per_second": 152079.74165767743, "data_wait_s": 1.105800038203597e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e12d1534d76c9ff900bbdeae0e7c252e3e1322b384410d78d3abc60e07050f5e
-size 714393
+oid sha256:dbe06dcf95b0f34ab5138aeeafe7e6ae129cac8e96c51e6aa25d2afaca246ffb
+size 716511