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": 75528.51987164348,
-      "required_pages_per_second_upper_estimate": 0.0065466206771905375
+      "required_token_positions_per_second": 75156.2396899193,
+      "required_pages_per_second_upper_estimate": 0.006514352374574179
     },
     {
       "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": 9441.064983955435,
-      "required_pages_per_second_upper_estimate": 0.0037661819786003812
+      "required_token_positions_per_second": 9394.529961239912,
+      "required_pages_per_second_upper_estimate": 0.0037476184622785673
     },
     {
       "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": 18882.12996791087,
-      "required_pages_per_second_upper_estimate": 0.004368048453938682
+      "required_token_positions_per_second": 18789.059922479824,
+      "required_pages_per_second_upper_estimate": 0.004346518336905065
     },
     {
       "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": 28323.194951866306,
-      "required_pages_per_second_upper_estimate": 0.0017247339722491835
+      "required_token_positions_per_second": 28183.589883719735,
+      "required_pages_per_second_upper_estimate": 0.0017162327560502429
     },
     {
       "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": 28323.194951866306,
-      "required_pages_per_second_upper_estimate": 0.00218039795101635
+      "required_token_positions_per_second": 28183.589883719735,
+      "required_pages_per_second_upper_estimate": 0.0021696507664188645
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28323.194951866306,
-      "required_pages_per_second_upper_estimate": 0.35167928966644285
+      "required_token_positions_per_second": 28183.589883719735,
+      "required_pages_per_second_upper_estimate": 0.34994586194816957
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,27 +158,34 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.03813167543752627,
-    "producer_p95_seconds": 0.13095372700081498,
-    "gpu_update_seconds": 0.9761610597598978,
-    "producer_batches_per_second": 26.22491638581075,
-    "gpu_batches_per_second": 1.0244211137104422,
-    "producer_headroom": 25.59974217128773,
+    "producer_batch_seconds": 0.08091448022655356,
+    "producer_p95_seconds": 0.18719405599949823,
+    "gpu_update_seconds": 0.9809963923712529,
+    "producer_batches_per_second": 12.35872735263313,
+    "gpu_batches_per_second": 1.019371740585928,
+    "producer_headroom": 12.123866947233024,
     "sustainable": true,
-    "buffer_seconds": 60.521985705113664,
+    "buffer_seconds": 60.82177632701768,
     "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": 0,
-    "upload_bytes_per_second": null,
-    "upload_sustainable": null,
+    "generation_bytes_per_second": 397083.56006381963,
+    "upload_bytes_per_second": 43958.880398765425,
+    "upload_sustainable": false,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 0.0,
+    "estimated_full_shards_per_hour": 85.20488835750524,
     "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": [],
+  "upload_measurements": [
+    {
+      "seconds": 0.9955667570011428,
+      "bytes": 43764,
+      "pages": 1,
+      "shards": 1
+    }
+  ],
   "cache_key": {
     "version": 1,
     "kind": "data",

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

@@ -47,3 +47,4 @@
 {"step": 89700, "loss": 0.5660558044910431, "lr": 0.0003, "grad_norm": 0.18036271697744438, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13127084061, "tokens_per_second": 149234.8550560003, "data_wait_s": 1.1087000530096702e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 89800, "loss": 0.6096045114099979, "lr": 0.0003, "grad_norm": 0.14591806689246076, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13141590162, "tokens_per_second": 148641.57574771694, "data_wait_s": 1.707599949440919e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 89900, "loss": 0.5985537357628345, "lr": 0.0003, "grad_norm": 0.12689724016327258, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13156147943, "tokens_per_second": 148430.52635961468, "data_wait_s": 2.1118999939062633e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 90000, "loss": 0.6308812461793423, "lr": 0.0003, "grad_norm": 0.13849090878893888, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13170804190, "tokens_per_second": 148330.89333950818, "data_wait_s": 7.241998901008628e-06, "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 sha256:78352c8bd175debd1407ff969c7f904257d1cf0e140f56e711b8f6c767718106
-size 103947
+oid sha256:0ea894cb7c733fdd0f407a434351b52862649be15f5e0da34f0d8ba60a6e0b1c
+size 106065