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": 75615.43433828405,
-      "required_pages_per_second_upper_estimate": 0.00655415420287624
+      "required_token_positions_per_second": 74102.64090433042,
+      "required_pages_per_second_upper_estimate": 0.006423029102161044
     },
     {
       "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": 9451.929292285506,
-      "required_pages_per_second_upper_estimate": 0.003770515913629131
+      "required_token_positions_per_second": 9262.830113041302,
+      "required_pages_per_second_upper_estimate": 0.0036950814237439373
     },
     {
       "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": 18903.85858457101,
-      "required_pages_per_second_upper_estimate": 0.004373074986992413
+      "required_token_positions_per_second": 18525.660226082604,
+      "required_pages_per_second_upper_estimate": 0.004285585452819897
     },
     {
       "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": 28355.787876856517,
-      "required_pages_per_second_upper_estimate": 0.0017267187103792256
+      "required_token_positions_per_second": 27788.490339123906,
+      "required_pages_per_second_upper_estimate": 0.0016921732667114632
     },
     {
       "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": 28355.787876856517,
-      "required_pages_per_second_upper_estimate": 0.0021829070446050874
+      "required_token_positions_per_second": 27788.490339123906,
+      "required_pages_per_second_upper_estimate": 0.002139234909770338
     },
     {
       "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": 28355.787876856517,
-      "required_pages_per_second_upper_estimate": 0.3520839847133183
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27788.490339123906,
+      "required_pages_per_second_upper_estimate": 0.3450400479174033
     }
   ],
   "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": 0.09713275687499845,
-    "producer_p95_seconds": 0.2761996490007732,
-    "gpu_update_seconds": 0.9750390333031727,
-    "producer_batches_per_second": 10.295188072206315,
-    "gpu_batches_per_second": 1.0255999666108404,
-    "producer_headroom": 10.0382102255984,
+    "producer_batch_seconds": 0.462495663257954,
+    "producer_p95_seconds": 0.2952149270022346,
+    "gpu_update_seconds": 0.9949442975343606,
+    "producer_batches_per_second": 2.162182436383747,
+    "gpu_batches_per_second": 1.0050813924740996,
+    "producer_headroom": 2.1512510853089597,
     "sustainable": true,
-    "buffer_seconds": 60.452420064796705,
+    "buffer_seconds": 60.691602149596,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 400745.3453038772,
-    "upload_bytes_per_second": 5258582.451440632,
+    "generation_bytes_per_second": 401734.17489576474,
+    "upload_bytes_per_second": 5247805.503957984,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.99062222802388,
+    "estimated_full_shards_per_hour": 86.20280203966814,
     "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": 1.212965857999734,
-      "bytes": 6163951,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.403984237998884,
       "bytes": 849560,
@@ -562,6 +556,12 @@
       "bytes": 2481076,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 1.3431768080008624,
+      "bytes": 5992564,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -212,3 +212,4 @@
 {"step": 176200, "loss": 0.60870729573071, "lr": 0.0003, "grad_norm": 0.1381184597474122, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25787859964, "tokens_per_second": 150172.95685653726, "data_wait_s": 1.4061999536352232e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 176300, "loss": 0.6374958213418722, "lr": 0.0003, "grad_norm": 0.1585988882770426, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25802571736, "tokens_per_second": 149971.39073706386, "data_wait_s": 1.3316999684320763e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 176400, "loss": 0.599961843341589, "lr": 0.0003, "grad_norm": 0.15021786931492198, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25817196045, "tokens_per_second": 133902.61068963964, "data_wait_s": 1.107099888031371e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 176500, "loss": 0.6475961171090603, "lr": 0.0003, "grad_norm": 0.18934323897787586, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25831661484, "tokens_per_second": 148787.94180501474, "data_wait_s": 1.8108999938704073e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:89293c85cfecd382f1be47628a3561662c7c0326bbdf40c3bfec1f87b0222dc2
-size 456549
+oid sha256:2d9691f7b87354a21971cb23f2fdc86984b4ff44fe44f7a259a5a9297ea8e1ef
+size 458667