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": 74723.14564520937,
-      "required_pages_per_second_upper_estimate": 0.006476812880445544
+      "required_token_positions_per_second": 75002.92959912904,
+      "required_pages_per_second_upper_estimate": 0.006501063844465353
     },
     {
       "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": 9340.393205651171,
-      "required_pages_per_second_upper_estimate": 0.003726022501057592
+      "required_token_positions_per_second": 9375.36619989113,
+      "required_pages_per_second_upper_estimate": 0.0037399737513527727
     },
     {
       "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": 18680.786411302342,
-      "required_pages_per_second_upper_estimate": 0.004321471165642869
+      "required_token_positions_per_second": 18750.73239978226,
+      "required_pages_per_second_upper_estimate": 0.004337651939070345
     },
     {
       "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": 28021.179616953516,
-      "required_pages_per_second_upper_estimate": 0.0017063428229049961
+      "required_token_positions_per_second": 28126.09859967339,
+      "required_pages_per_second_upper_estimate": 0.0017127318384852765
     },
     {
       "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": 28021.179616953516,
-      "required_pages_per_second_upper_estimate": 0.002157147974502808
+      "required_token_positions_per_second": 28126.09859967339,
+      "required_pages_per_second_upper_estimate": 0.0021652249282269167
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28021.179616953516,
-      "required_pages_per_second_upper_estimate": 0.3479292699871303
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28126.09859967339,
+      "required_pages_per_second_upper_estimate": 0.3492320126112643
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.08991390005469668,
-    "producer_p95_seconds": 0.19443707799928234,
-    "gpu_update_seconds": 0.9866822302966956,
-    "producer_batches_per_second": 11.121750912725142,
-    "gpu_batches_per_second": 1.013497526654858,
-    "producer_headroom": 10.973633995371951,
+    "producer_batch_seconds": 0.3064620796797044,
+    "producer_p95_seconds": 0.19963705000009213,
+    "gpu_update_seconds": 0.9830016026581467,
+    "producer_batches_per_second": 3.263046446219837,
+    "gpu_batches_per_second": 1.0172923393979092,
+    "producer_headroom": 3.20757988618207,
     "sustainable": true,
-    "buffer_seconds": 60.18761604809843,
+    "buffer_seconds": 60.9460993648051,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 466120.28843024524,
-    "upload_bytes_per_second": 5377124.927694322,
+    "generation_bytes_per_second": 448721.47799403203,
+    "upload_bytes_per_second": 5315900.009096232,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 100.0185631721546,
+    "estimated_full_shards_per_hour": 96.28518347612115,
     "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": [
@@ -520,6 +520,12 @@
       "bytes": 6054797,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.2160326559996975,
+      "bytes": 2369678,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -68,3 +68,4 @@
 {"step": 321800, "loss": 0.5576529204845428, "lr": 0.0003, "grad_norm": 0.16727996310470222, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47102413094, "tokens_per_second": 150722.54404603963, "data_wait_s": 1.4507999367197044e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 321900, "loss": 0.6707786060869694, "lr": 0.0003, "grad_norm": 0.1912697240698255, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47117139175, "tokens_per_second": 149272.51686611003, "data_wait_s": 1.2696999874606263e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 322000, "loss": 0.642135227099061, "lr": 0.0003, "grad_norm": 0.18308558477457074, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47131682588, "tokens_per_second": 148355.99669807407, "data_wait_s": 1.3996000234328676e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 322100, "loss": 0.5703791789710522, "lr": 0.0003, "grad_norm": 0.1460847661649317, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47146211947, "tokens_per_second": 148092.0903678938, "data_wait_s": 1.1665999409160577e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6b98c62b22b8d9954a466492add1ef857d5bf6f488579cfcb05371f71db31b10
-size 149208
+oid sha256:7e8a44368e05400e763e71f74595f3ebce038b5e67c45db3bb510b83c6724afa
+size 151326