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": 75206.94243775622,
-      "required_pages_per_second_upper_estimate": 0.006518747160251745
+      "required_token_positions_per_second": 74661.84648542074,
+      "required_pages_per_second_upper_estimate": 0.0064714996246363655
     },
     {
       "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": 9400.867804719528,
-      "required_pages_per_second_upper_estimate": 0.0037501467228017904
+      "required_token_positions_per_second": 9332.730810677593,
+      "required_pages_per_second_upper_estimate": 0.0037229658571396175
     },
     {
       "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": 18801.735609439056,
-      "required_pages_per_second_upper_estimate": 0.004349450633998612
+      "required_token_positions_per_second": 18665.461621355185,
+      "required_pages_per_second_upper_estimate": 0.0043179260451956854
     },
     {
       "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": 28202.603414158588,
-      "required_pages_per_second_upper_estimate": 0.0017173905802976845
+      "required_token_positions_per_second": 27998.192432032778,
+      "required_pages_per_second_upper_estimate": 0.0017049430239477617
     },
     {
       "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": 28202.603414158588,
-      "required_pages_per_second_upper_estimate": 0.002171114480624868
+      "required_token_positions_per_second": 27998.192432032778,
+      "required_pages_per_second_upper_estimate": 0.0021553783573750093
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28202.603414158588,
-      "required_pages_per_second_upper_estimate": 0.35018194636202726
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27998.192432032778,
+      "required_pages_per_second_upper_estimate": 0.34764384608357374
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.09352632367972546,
-    "producer_p95_seconds": 0.24202358600086882,
-    "gpu_update_seconds": 0.9803350277272573,
-    "producer_batches_per_second": 10.692176925765114,
-    "gpu_batches_per_second": 1.0200594406162684,
-    "producer_headroom": 10.481915562984684,
+    "producer_batch_seconds": 0.3960835619999443,
+    "producer_p95_seconds": 0.29465220400015824,
+    "gpu_update_seconds": 0.9874923199816241,
+    "producer_batches_per_second": 2.524719771127842,
+    "gpu_batches_per_second": 1.0126661035891487,
+    "producer_headroom": 2.4931413840945083,
     "sustainable": true,
-    "buffer_seconds": 60.780771719089955,
+    "buffer_seconds": 60.23703151887907,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 414028.5245373916,
-    "upload_bytes_per_second": 5920925.7460016385,
+    "generation_bytes_per_second": 369410.50482843834,
+    "upload_bytes_per_second": 5912253.735729597,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 88.84088327494918,
+    "estimated_full_shards_per_hour": 79.26689489974845,
     "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": 1.1174536390008143,
-      "bytes": 7647354,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.2463584760007507,
       "bytes": 4721254,
@@ -562,6 +556,12 @@
       "bytes": 2655724,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.8924994240005617,
+      "bytes": 5699054,
+      "pages": 19,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -145,3 +145,4 @@
 {"step": 99500, "loss": 0.6166529804468155, "lr": 0.0003, "grad_norm": 0.13084167545458994, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14560117029, "tokens_per_second": 148304.48641367702, "data_wait_s": 1.5222998626995832e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 99600, "loss": 0.6056494414806366, "lr": 0.0003, "grad_norm": 0.1452663709025324, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14574631938, "tokens_per_second": 147439.03541988644, "data_wait_s": 1.087899727281183e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 99700, "loss": 0.5871254727244377, "lr": 0.0003, "grad_norm": 0.14458459120061495, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14589058669, "tokens_per_second": 137545.56160873885, "data_wait_s": 1.1546999303391203e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 99800, "loss": 0.643492616713047, "lr": 0.0003, "grad_norm": 0.16165238422953543, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14603748246, "tokens_per_second": 149025.5432800379, "data_wait_s": 1.6723999578971416e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c8ab19e0b3ae2db8482f8e9493d2f14a2ae49bcefbc97971dea67a849d4701d5
-size 313077
+oid sha256:876c2ace1ded5a0d482ca8ee95f2dd6dca10125abbe584cf2ceade0f0a3646a7
+size 315195