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": 77392.59729152752,
-      "required_pages_per_second_upper_estimate": 0.005354199456194275
+      "required_token_positions_per_second": 77395.2608738646,
+      "required_pages_per_second_upper_estimate": 0.0053543837290007164
     },
     {
       "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": 9674.07466144094,
-      "required_pages_per_second_upper_estimate": 0.004050333043094198
+      "required_token_positions_per_second": 9674.407609233074,
+      "required_pages_per_second_upper_estimate": 0.004050472441382024
     },
     {
       "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": 19348.14932288188,
-      "required_pages_per_second_upper_estimate": 0.004029709258946161
+      "required_token_positions_per_second": 19348.81521846615,
+      "required_pages_per_second_upper_estimate": 0.004029847947435516
     },
     {
       "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": 29022.22398432282,
-      "required_pages_per_second_upper_estimate": 0.0017398380134900143
+      "required_token_positions_per_second": 29023.22282769922,
+      "required_pages_per_second_upper_estimate": 0.001739897892625284
     },
     {
       "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": 29022.22398432282,
-      "required_pages_per_second_upper_estimate": 0.002256350259508817
+      "required_token_positions_per_second": 29023.22282769922,
+      "required_pages_per_second_upper_estimate": 0.0022564279151878848
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 29022.22398432282,
-      "required_pages_per_second_upper_estimate": 0.3722659276346227
+      "required_token_positions_per_second": 29023.22282769922,
+      "required_pages_per_second_upper_estimate": 0.3722787397249807
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.06360460313278082,
-    "producer_p95_seconds": 0.15299785299976065,
-    "gpu_update_seconds": 0.9526492530322574,
-    "producer_batches_per_second": 15.722132530445984,
-    "gpu_batches_per_second": 1.0497042818403797,
-    "producer_headroom": 14.977677811203524,
+    "producer_batch_seconds": 0.04820530507033283,
+    "producer_p95_seconds": 0.14429516400014109,
+    "gpu_update_seconds": 0.9526164673074579,
+    "producer_batches_per_second": 20.74460473885547,
+    "gpu_batches_per_second": 1.0497404089879638,
+    "producer_headroom": 19.76165208201805,
     "sustainable": true,
-    "buffer_seconds": 60.01690294103222,
+    "buffer_seconds": 60.014837440369845,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 472239.58878109255,
-    "upload_bytes_per_second": 5382884.021252117,
+    "generation_bytes_per_second": 503169.68527313176,
+    "upload_bytes_per_second": 5422011.182122283,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 101.33162257742484,
+    "estimated_full_shards_per_hour": 107.96850126882043,
     "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": [
@@ -346,6 +346,12 @@
       "bytes": 2549967,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.9600205120000282,
+      "bytes": 6407325,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -31,3 +31,4 @@
 {"step": 560200, "loss": 0.5885102152824402, "lr": 0.0003, "grad_norm": 0.1723210634054678, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81995415301, "tokens_per_second": 153730.4404726955, "data_wait_s": 1.071900032911799e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 560300, "loss": 0.6503345370292664, "lr": 0.0003, "grad_norm": 0.1570713956406031, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82010105414, "tokens_per_second": 153278.1151601721, "data_wait_s": 1.6234000213444233e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 560400, "loss": 0.5924027562141418, "lr": 0.0003, "grad_norm": 0.15323526764925785, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82024829166, "tokens_per_second": 153491.6710915769, "data_wait_s": 1.574399993842235e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 560500, "loss": 0.5484539866447449, "lr": 0.0003, "grad_norm": 0.15213750671061338, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82039572408, "tokens_per_second": 153307.04067925774, "data_wait_s": 1.7552999906911282e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ee1482225ac24e60e8c9fea5009b34ad8799a66d847e3aa17d59de145f46e97c
-size 70842
+oid sha256:21b85ff74efa59657cc0b996fecbff1f46d81980700f3ffc75cf274e972a3cad
+size 72960