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": 77538.96167812563,
-      "required_pages_per_second_upper_estimate": 0.005364325284071299
+      "required_token_positions_per_second": 77348.79866993793,
+      "required_pages_per_second_upper_estimate": 0.0053511693659259035
     },
     {
       "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": 9692.370209765704,
-      "required_pages_per_second_upper_estimate": 0.004057993007123283
+      "required_token_positions_per_second": 9668.59983374224,
+      "required_pages_per_second_upper_estimate": 0.004048040847064155
     },
     {
       "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": 19384.74041953141,
-      "required_pages_per_second_upper_estimate": 0.004037330219406147
+      "required_token_positions_per_second": 19337.19966748448,
+      "required_pages_per_second_upper_estimate": 0.004027428734488713
     },
     {
       "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": 29077.110629297113,
-      "required_pages_per_second_upper_estimate": 0.0017431283840491688
+      "required_token_positions_per_second": 29005.799501226724,
+      "required_pages_per_second_upper_estimate": 0.001738853390807137
     },
     {
       "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": 29077.110629297113,
-      "required_pages_per_second_upper_estimate": 0.0022606174547347316
+      "required_token_positions_per_second": 29005.799501226724,
+      "required_pages_per_second_upper_estimate": 0.002255073328191762
     },
     {
       "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": 29077.110629297113,
-      "required_pages_per_second_upper_estimate": 0.37296995458366505
+      "required_token_positions_per_second": 29005.799501226724,
+      "required_pages_per_second_upper_estimate": 0.37205525200070194
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.05347744585928638,
-    "producer_p95_seconds": 0.18114154999966559,
-    "gpu_update_seconds": 0.9508510096647227,
-    "producer_batches_per_second": 18.69947197237636,
-    "gpu_batches_per_second": 1.0516894758860356,
-    "producer_headroom": 17.780411805131248,
+    "producer_batch_seconds": 0.09741961384362696,
+    "producer_p95_seconds": 0.18321653400016658,
+    "gpu_update_seconds": 0.9531886890009945,
+    "producer_batches_per_second": 10.264873371446015,
+    "gpu_batches_per_second": 1.049110225015434,
+    "producer_headroom": 9.784361191689847,
     "sustainable": true,
-    "buffer_seconds": 60.85446461854225,
+    "buffer_seconds": 60.05088740706265,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 310233.50192612875,
-    "upload_bytes_per_second": 5706639.795803394,
+    "generation_bytes_per_second": 347508.2062114373,
+    "upload_bytes_per_second": 5701009.005794727,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.56888764703652,
+    "estimated_full_shards_per_hour": 74.5671714759573,
     "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": 0.8888624019999725,
-      "bytes": 6324210,
-      "pages": 34,
-      "shards": 2
-    },
     {
       "seconds": 1.1897545489999857,
       "bytes": 1226596,
@@ -562,6 +556,12 @@
       "bytes": 12042262,
       "pages": 38,
       "shards": 4
+    },
+    {
+      "seconds": 1.2374154409990297,
+      "bytes": 7885146,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -81,3 +81,4 @@
 {"step": 565200, "loss": 0.6330256462097168, "lr": 0.0003, "grad_norm": 0.16335342735842637, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82728290587, "tokens_per_second": 154213.20827823642, "data_wait_s": 7.676000677747652e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 565300, "loss": 0.5858858823776245, "lr": 0.0003, "grad_norm": 0.17481800118536958, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82742866371, "tokens_per_second": 152102.93307452483, "data_wait_s": 8.957000318332575e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 565400, "loss": 0.5888734459877014, "lr": 0.0003, "grad_norm": 0.16529429363687742, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82757312201, "tokens_per_second": 151146.1973423909, "data_wait_s": 1.558599979034625e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 565500, "loss": 0.5776864290237427, "lr": 0.0003, "grad_norm": 0.14738988748703424, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82771885314, "tokens_per_second": 151557.58958653844, "data_wait_s": 1.4744000509381294e-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 sha256:09cb48ab43acaffca3492038d7d59a97283184bf7d92aeabc8d8b599d844c0ae
-size 177525
+oid sha256:2c6b56cf2756d24f373ba4813880572b97ca9ea5a3a848f102777feb98e6ff2f
+size 179643