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": 75548.50040161255,
-      "required_pages_per_second_upper_estimate": 0.005226620554455656
+      "required_token_positions_per_second": 77031.43992682225,
+      "required_pages_per_second_upper_estimate": 0.0053292137517034245
     },
     {
       "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": 9443.562550201568,
-      "required_pages_per_second_upper_estimate": 0.003953822435758533
+      "required_token_positions_per_second": 9628.929990852781,
+      "required_pages_per_second_upper_estimate": 0.004031431912246859
     },
     {
       "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": 18887.125100403136,
-      "required_pages_per_second_upper_estimate": 0.003933690071430177
+      "required_token_positions_per_second": 19257.859981705562,
+      "required_pages_per_second_upper_estimate": 0.004010904370269182
     },
     {
       "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": 28330.687650604705,
-      "required_pages_per_second_upper_estimate": 0.0016983814661984554
+      "required_token_positions_per_second": 28886.78997255834,
+      "required_pages_per_second_upper_estimate": 0.001731718951280496
     },
     {
       "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": 28330.687650604705,
-      "required_pages_per_second_upper_estimate": 0.002202586351309104
+      "required_token_positions_per_second": 28886.78997255834,
+      "required_pages_per_second_upper_estimate": 0.00224582086080539
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28330.687650604705,
-      "required_pages_per_second_upper_estimate": 0.3633956420595516
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28886.78997255834,
+      "required_pages_per_second_upper_estimate": 0.3705287255494201
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10251519771105677,
-    "producer_p95_seconds": 0.25733189200036577,
-    "gpu_update_seconds": 0.9759028916267716,
-    "producer_batches_per_second": 9.75465123540551,
-    "gpu_batches_per_second": 1.0246921169923577,
-    "producer_headroom": 9.519592347442897,
+    "producer_batch_seconds": 0.09022814403917323,
+    "producer_p95_seconds": 0.24863629799801856,
+    "gpu_update_seconds": 0.9571156928916242,
+    "producer_batches_per_second": 11.083016398584487,
+    "gpu_batches_per_second": 1.0448057715769077,
+    "producer_headroom": 10.607728919660424,
     "sustainable": true,
-    "buffer_seconds": 60.50597928085984,
+    "buffer_seconds": 60.298288652172324,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 156,
+    "upload_pages_per_commit": 161,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 675210.7236299671,
-    "upload_bytes_per_second": 5038855.982727191,
+    "generation_bytes_per_second": 674620.0493191228,
+    "upload_bytes_per_second": 5013461.935723811,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.44225159489756,
+    "estimated_full_shards_per_hour": 72.37887911644107,
     "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": 1.4409255560003658,
-      "bytes": 6062763,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.1668520560015168,
       "bytes": 7696896,
@@ -562,6 +556,12 @@
       "bytes": 16328441,
       "pages": 37,
       "shards": 4
+    },
+    {
+      "seconds": 1.2096843409999565,
+      "bytes": 2710723,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -210,3 +210,4 @@
 {"step": 906000, "loss": 0.534150242805481, "lr": 3e-05, "grad_norm": 0.33707402206018144, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132612582574, "tokens_per_second": 152527.3856317226, "data_wait_s": 1.629799953661859e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 906100, "loss": 0.4887656271457672, "lr": 3e-05, "grad_norm": 0.3205562139009218, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132627298243, "tokens_per_second": 152940.06555462346, "data_wait_s": 1.747599890222773e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 906200, "loss": 0.5246565937995911, "lr": 3e-05, "grad_norm": 0.3291295920015878, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132641939794, "tokens_per_second": 152452.25118830553, "data_wait_s": 1.3370998203754425e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 906300, "loss": 0.5036863684654236, "lr": 3e-05, "grad_norm": 0.32663316570053796, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132656594751, "tokens_per_second": 152726.03360806828, "data_wait_s": 1.52709981193766e-05, "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 sha256:0e0e14ec27196d4a0d388a78ef22e881d8202bb9f4e97ecd669e172bfabec1fd
-size 452313
+oid sha256:2e801825d1f085f6e5ba642209ff4fbaa8d45b64039068310fe3de3844a27ebe
+size 454431