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": 75131.69605351688,
-      "required_pages_per_second_upper_estimate": 0.006512224994376095
+      "required_token_positions_per_second": 74925.11346832884,
+      "required_pages_per_second_upper_estimate": 0.006494318939470772
     },
     {
       "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": 9391.46200668961,
-      "required_pages_per_second_upper_estimate": 0.003746394609338443
+      "required_token_positions_per_second": 9365.639183541105,
+      "required_pages_per_second_upper_estimate": 0.003736093499098893
     },
     {
       "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": 18782.92401337922,
-      "required_pages_per_second_upper_estimate": 0.004345098902322952
+      "required_token_positions_per_second": 18731.27836708221,
+      "required_pages_per_second_upper_estimate": 0.00433315159098447
     },
     {
       "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": 28174.38602006883,
-      "required_pages_per_second_upper_estimate": 0.0017156722890428434
+      "required_token_positions_per_second": 28096.91755062331,
+      "required_pages_per_second_upper_estimate": 0.0017109548656992588
     },
     {
       "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": 28174.38602006883,
-      "required_pages_per_second_upper_estimate": 0.0021689422275170925
+      "required_token_positions_per_second": 28096.91755062331,
+      "required_pages_per_second_upper_estimate": 0.0021629784902927277
     },
     {
       "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": 28174.38602006883,
-      "required_pages_per_second_upper_estimate": 0.34983158076497545
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28096.91755062331,
+      "required_pages_per_second_upper_estimate": 0.348869681644751
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.1616476905312254,
-    "producer_p95_seconds": 0.3656139640006586,
-    "gpu_update_seconds": 0.9813168592318612,
-    "producer_batches_per_second": 6.186293145999698,
-    "gpu_batches_per_second": 1.0190388462119802,
-    "producer_headroom": 6.070713760320013,
+    "producer_batch_seconds": 0.534133971390645,
+    "producer_p95_seconds": 0.3778401669987943,
+    "gpu_update_seconds": 0.9840225337952293,
+    "producer_batches_per_second": 1.8721894759781879,
+    "gpu_batches_per_second": 1.0162368905752066,
+    "producer_headroom": 1.842276631896819,
     "sustainable": true,
-    "buffer_seconds": 60.841645272375395,
+    "buffer_seconds": 60.02537456150899,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 296352.6882195805,
-    "upload_bytes_per_second": 3621719.222119112,
+    "generation_bytes_per_second": 268330.06452815473,
+    "upload_bytes_per_second": 3656833.3553741463,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.59038815441667,
+    "estimated_full_shards_per_hour": 57.577385443529906,
     "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": 0.927039637001144,
-      "bytes": 2682847,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 0.8612236650005798,
       "bytes": 6230072,
@@ -562,6 +556,12 @@
       "bytes": 6227883,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.9593730830019922,
+      "bytes": 6813129,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -312,3 +312,4 @@
 {"step": 71200, "loss": 0.6313820295035839, "lr": 0.0003, "grad_norm": 0.14232023127594137, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10420259971, "tokens_per_second": 150352.63387042785, "data_wait_s": 1.6206999134737998e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 71300, "loss": 0.6463793590664864, "lr": 0.0003, "grad_norm": 0.14522967841544823, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10434922937, "tokens_per_second": 149927.87483328488, "data_wait_s": 1.0823001503013074e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 71400, "loss": 0.6088871471583843, "lr": 0.0003, "grad_norm": 0.16187098194106173, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10449669799, "tokens_per_second": 149136.1061377488, "data_wait_s": 1.0667998139979318e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 71500, "loss": 0.5960981119424105, "lr": 0.0003, "grad_norm": 0.13603921606396208, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10464120284, "tokens_per_second": 147566.92890491622, "data_wait_s": 8.75100158737041e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4051091a72257ab4d8af872d16efcb2aa48fe2c088389ea0eacef5d37b877c56
-size 669915
+oid sha256:d2854d935221d42084829ba684bffb0f795e5babe94935ea2c01b2404b132f93
+size 672033