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": 75316.0245631188,
-      "required_pages_per_second_upper_estimate": 0.005210537336530396
+      "required_token_positions_per_second": 76414.71866325618,
+      "required_pages_per_second_upper_estimate": 0.0052865475436994405
     },
     {
       "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": 9414.50307038985,
-      "required_pages_per_second_upper_estimate": 0.00394165583839231
+      "required_token_positions_per_second": 9551.839832907022,
+      "required_pages_per_second_upper_estimate": 0.003999155872940527
     },
     {
       "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": 18829.0061407797,
-      "required_pages_per_second_upper_estimate": 0.00392158542484065
+      "required_token_positions_per_second": 19103.679665814045,
+      "required_pages_per_second_upper_estimate": 0.003978792676477336
     },
     {
       "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": 28243.50921116955,
-      "required_pages_per_second_upper_estimate": 0.0016931552518680872
+      "required_token_positions_per_second": 28655.519498721067,
+      "required_pages_per_second_upper_estimate": 0.001717854639503518
     },
     {
       "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": 28243.50921116955,
-      "required_pages_per_second_upper_estimate": 0.0021958086111004507
+      "required_token_positions_per_second": 28655.519498721067,
+      "required_pages_per_second_upper_estimate": 0.00222784059871585
     },
     {
       "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": 28243.50921116955,
-      "required_pages_per_second_upper_estimate": 0.3622774106433928
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28655.519498721067,
+      "required_pages_per_second_upper_estimate": 0.3675622362299235
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 0.31428951145309725,
-    "producer_p95_seconds": 0.14616541999976107,
-    "gpu_update_seconds": 0.9789151834243728,
-    "producer_batches_per_second": 3.1817797398855743,
-    "gpu_batches_per_second": 1.021538961630843,
-    "producer_headroom": 3.11469249768604,
+    "producer_batch_seconds": 0.0946680697187432,
+    "producer_p95_seconds": 0.15241029299977527,
+    "gpu_update_seconds": 0.964840298959995,
+    "producer_batches_per_second": 10.563223724440338,
+    "gpu_batches_per_second": 1.0364409540914739,
+    "producer_headroom": 10.191823936270326,
     "sustainable": true,
-    "buffer_seconds": 60.69274137231111,
+    "buffer_seconds": 60.78493883447969,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 295280.84868495096,
-    "upload_bytes_per_second": 5337662.209202704,
+    "generation_bytes_per_second": 322945.7918455066,
+    "upload_bytes_per_second": 6185064.992552971,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.36039634143254,
+    "estimated_full_shards_per_hour": 69.29664913677118,
     "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": [
@@ -220,6 +220,12 @@
       "bytes": 9808086,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.0871454139996786,
+      "bytes": 13468906,
+      "pages": 38,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791033297667653750-ba4975eb/train.jsonl

@@ -19,3 +19,4 @@
 {"step": 601900, "loss": 0.6145690083503723, "lr": 0.0003, "grad_norm": 0.16814787770286502, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88100145606, "tokens_per_second": 152749.24062053088, "data_wait_s": 1.8597999769554008e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 602000, "loss": 0.5783024430274963, "lr": 0.0003, "grad_norm": 0.1514827487704976, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88114623848, "tokens_per_second": 147929.7346813375, "data_wait_s": 1.7145000128948595e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 602100, "loss": 0.576166033744812, "lr": 0.0003, "grad_norm": 0.17736680918012135, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88129104162, "tokens_per_second": 150394.35689263276, "data_wait_s": 1.9164000150340144e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 602200, "loss": 0.6551995873451233, "lr": 0.0003, "grad_norm": 0.1718665374166175, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88143818424, "tokens_per_second": 151097.07521291965, "data_wait_s": 1.70569996953418e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791033297.83ac8725-d818-41af-bbb1-78dcd88792cf-7zwhs.737.0.1791033297667653750-ba4975eb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1b69fc3c9944a53d0de23a481fb03e7870b59f4896c700b3bf3d5ba9d8bb20ee
-size 44643
+oid sha256:17e36e87ad10fd740a61a688894983b7f56a70b84b32228016092a322bcd7b0f
+size 46761