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": 77280.73115412582,
-      "required_pages_per_second_upper_estimate": 0.0053464602972436255
+      "required_token_positions_per_second": 77582.35473528446,
+      "required_pages_per_second_upper_estimate": 0.005367327316451311
     },
     {
       "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": 9660.091394265728,
-      "required_pages_per_second_upper_estimate": 0.00404447854113176
+      "required_token_positions_per_second": 9697.794341910558,
+      "required_pages_per_second_upper_estimate": 0.004060263977983573
     },
     {
       "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": 19320.182788531456,
-      "required_pages_per_second_upper_estimate": 0.0040238845673680745
+      "required_token_positions_per_second": 19395.588683821115,
+      "required_pages_per_second_upper_estimate": 0.004039589626769725
     },
     {
       "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": 28980.274182797188,
-      "required_pages_per_second_upper_estimate": 0.0017373231869421907
+      "required_token_positions_per_second": 29093.38302573167,
+      "required_pages_per_second_upper_estimate": 0.0017441038893689098
     },
     {
       "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": 28980.274182797188,
-      "required_pages_per_second_upper_estimate": 0.0022530888469578757
+      "required_token_positions_per_second": 29093.38302573167,
+      "required_pages_per_second_upper_estimate": 0.0022618825619828108
     },
     {
       "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": 28980.274182797188,
-      "required_pages_per_second_upper_estimate": 0.3717278406228395
+      "required_token_positions_per_second": 29093.38302573167,
+      "required_pages_per_second_upper_estimate": 0.37317867941318955
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.23549360834378064,
-    "producer_p95_seconds": 0.1584173479996025,
-    "gpu_update_seconds": 0.9540282409202315,
-    "producer_batches_per_second": 4.2463997517086325,
-    "gpu_batches_per_second": 1.0481870002458473,
-    "producer_headroom": 4.051185285366694,
+    "producer_batch_seconds": 0.22235327151563666,
+    "producer_p95_seconds": 0.16241001499929553,
+    "gpu_update_seconds": 0.9503191834221101,
+    "producer_batches_per_second": 4.497347815859217,
+    "gpu_batches_per_second": 1.0522780318913365,
+    "producer_headroom": 4.273915903932542,
     "sustainable": true,
-    "buffer_seconds": 60.10377917797459,
+    "buffer_seconds": 60.820427739015045,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 486911.6116432644,
-    "upload_bytes_per_second": 3768783.117588168,
+    "generation_bytes_per_second": 485630.3137384657,
+    "upload_bytes_per_second": 3848852.41133768,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 104.47989713643503,
+    "estimated_full_shards_per_hour": 104.2049604331539,
     "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": [
@@ -394,6 +394,12 @@
       "bytes": 7420139,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.4091035430001284,
+      "bytes": 10556108,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -46,3 +46,4 @@
 {"step": 444800, "loss": 0.5636985301971436, "lr": 0.0003, "grad_norm": 0.16352203214652344, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65105497414, "tokens_per_second": 151367.44674651648, "data_wait_s": 1.705900012893835e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 444900, "loss": 0.5851013660430908, "lr": 0.0003, "grad_norm": 0.16064967921121015, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65119922459, "tokens_per_second": 149715.8220490589, "data_wait_s": 7.629000720044132e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 445000, "loss": 0.6301952004432678, "lr": 0.0003, "grad_norm": 0.20991914998299785, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65134552525, "tokens_per_second": 152765.13751353952, "data_wait_s": 1.3879000107408501e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 445100, "loss": 0.6043710708618164, "lr": 0.0003, "grad_norm": 0.14340291641602193, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65149310375, "tokens_per_second": 139345.12325619237, "data_wait_s": 1.451499974791659e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1a5e8d60f550b51319ec55c2ec400ec668a3819671f016bec542e3c97563b958
-size 102612
+oid sha256:5817175f6ef6fd27f97fc5ece4d50e98a581085cb3ff6240d4747e8586fa94cb
+size 104730