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": 75804.54427841451,
-      "required_pages_per_second_upper_estimate": 0.0065705457731919875
+      "required_token_positions_per_second": 74535.33057357532,
+      "required_pages_per_second_upper_estimate": 0.006460533546049252
     },
     {
       "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": 9475.568034801814,
-      "required_pages_per_second_upper_estimate": 0.003779945761449583
+      "required_token_positions_per_second": 9316.916321696915,
+      "required_pages_per_second_upper_estimate": 0.003716657221037544
     },
     {
       "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": 18951.136069603628,
-      "required_pages_per_second_upper_estimate": 0.004384011801099391
+      "required_token_positions_per_second": 18633.83264339383,
+      "required_pages_per_second_upper_estimate": 0.004310609237795326
     },
     {
       "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": 28426.704104405446,
-      "required_pages_per_second_upper_estimate": 0.0017310371365682644
+      "required_token_positions_per_second": 27950.74896509075,
+      "required_pages_per_second_upper_estimate": 0.0017020539657276257
     },
     {
       "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": 28426.704104405446,
-      "required_pages_per_second_upper_estimate": 0.0021883663721104863
+      "required_token_positions_per_second": 27950.74896509075,
+      "required_pages_per_second_upper_estimate": 0.0021517260279578895
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28426.704104405446,
-      "required_pages_per_second_upper_estimate": 0.35296452691813013
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27950.74896509075,
+      "required_pages_per_second_upper_estimate": 0.3470547570072234
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.11064865784385347,
-    "producer_p95_seconds": 0.3781688369999756,
-    "gpu_update_seconds": 0.9726065990082626,
-    "producer_batches_per_second": 9.037615272398444,
-    "gpu_batches_per_second": 1.0281649343317942,
-    "producer_headroom": 8.790044253232583,
+    "producer_batch_seconds": 0.14013973979706407,
+    "producer_p95_seconds": 0.35657428500053356,
+    "gpu_update_seconds": 0.9891684846989658,
+    "producer_batches_per_second": 7.135734670608758,
+    "gpu_batches_per_second": 1.0109501217119048,
+    "producer_headroom": 7.058443851339939,
     "sustainable": true,
-    "buffer_seconds": 60.30160913851228,
+    "buffer_seconds": 60.339277566636916,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 394286.40371858043,
-    "upload_bytes_per_second": 3297555.4995574937,
+    "generation_bytes_per_second": 336354.4082880705,
+    "upload_bytes_per_second": 3232662.15770825,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.60468372028407,
+    "estimated_full_shards_per_hour": 72.17382608872973,
     "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": 1.2521359960010159,
-      "bytes": 10213095,
-      "pages": 25,
-      "shards": 3
-    },
     {
       "seconds": 1.2234030319996236,
       "bytes": 7751215,
@@ -562,6 +556,12 @@
       "bytes": 5944089,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 0.9463060489979398,
+      "bytes": 743440,
+      "pages": 25,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -320,3 +320,4 @@
 {"step": 347000, "loss": 0.6010876540094614, "lr": 0.0003, "grad_norm": 0.15745685106639337, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50792017038, "tokens_per_second": 150406.42502960155, "data_wait_s": 1.6705998859833926e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 347100, "loss": 0.5362796876579523, "lr": 0.0003, "grad_norm": 0.1659641151836233, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50806674727, "tokens_per_second": 149587.77980125116, "data_wait_s": 8.677001460455358e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 347200, "loss": 0.5770402681082487, "lr": 0.0003, "grad_norm": 0.15166628420578193, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50821129195, "tokens_per_second": 146713.61022394153, "data_wait_s": 1.186599547509104e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 347300, "loss": 0.5554037876427174, "lr": 0.0003, "grad_norm": 0.19573400654013476, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50835576533, "tokens_per_second": 147477.54933544682, "data_wait_s": 1.2800002878066152e-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:d586c411d1547daec45675ded2eecc07de8d88d0c309b7fc3ba9fff24a837e1b
-size 686859
+oid sha256:66ba90d1a6f4e8aa6a2e01ef21cd5f8501ff9c95b600a8968867fc2ee5ac4a2e
+size 688977