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": 76101.41323470626,
-      "required_pages_per_second_upper_estimate": 0.005264872347183605
+      "required_token_positions_per_second": 76287.25652997452,
+      "required_pages_per_second_upper_estimate": 0.005277729417566119
     },
     {
       "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": 9512.676654338282,
-      "required_pages_per_second_upper_estimate": 0.003982759067893961
+      "required_token_positions_per_second": 9535.907066246815,
+      "required_pages_per_second_upper_estimate": 0.0039924851562539
     },
     {
       "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": 19025.353308676564,
-      "required_pages_per_second_upper_estimate": 0.003962479361890542
+      "required_token_positions_per_second": 19071.81413249363,
+      "required_pages_per_second_upper_estimate": 0.003972155926237318
     },
     {
       "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": 28538.029963014844,
-      "required_pages_per_second_upper_estimate": 0.0017108113212340635
+      "required_token_positions_per_second": 28607.721198740444,
+      "required_pages_per_second_upper_estimate": 0.0017149892044034578
     },
     {
       "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": 28538.029963014844,
-      "required_pages_per_second_upper_estimate": 0.0022187063040965435
+      "required_token_positions_per_second": 28607.721198740444,
+      "required_pages_per_second_upper_estimate": 0.002224124491134859
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28538.029963014844,
-      "required_pages_per_second_upper_estimate": 0.36605520661631896
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28607.721198740444,
+      "required_pages_per_second_upper_estimate": 0.36694913095958803
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.44608682596864924,
-    "producer_p95_seconds": 0.28946338900277624,
-    "gpu_update_seconds": 0.9688124946196945,
-    "producer_batches_per_second": 2.2417160556771956,
-    "gpu_batches_per_second": 1.0321914772502476,
-    "producer_headroom": 2.171802524129646,
+    "producer_batch_seconds": 0.09511008100781737,
+    "producer_p95_seconds": 0.2706608199987386,
+    "gpu_update_seconds": 0.9664523716491372,
+    "producer_batches_per_second": 10.514132565167378,
+    "gpu_batches_per_second": 1.0347121382646283,
+    "producer_headroom": 10.16140835343944,
     "sustainable": true,
-    "buffer_seconds": 60.06637466642106,
+    "buffer_seconds": 60.88649941389564,
     "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": 356073.0591577522,
-    "upload_bytes_per_second": 3232194.2676957883,
+    "generation_bytes_per_second": 370991.5661673316,
+    "upload_bytes_per_second": 3229219.904543384,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.4049895386641,
+    "estimated_full_shards_per_hour": 79.60615385785066,
     "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": 1.3335332679998828,
-      "bytes": 5896707,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.9101363129993842,
       "bytes": 4340262,
@@ -562,6 +556,12 @@
       "bytes": 5921876,
       "pages": 24,
       "shards": 2
+    },
+    {
+      "seconds": 1.7577662710027653,
+      "bytes": 6866333,
+      "pages": 46,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -281,3 +281,4 @@
 {"step": 828100, "loss": 0.5479229092597961, "lr": 3e-05, "grad_norm": 0.2848581981503589, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121210710277, "tokens_per_second": 151588.3865879714, "data_wait_s": 8.810002327663824e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 828200, "loss": 0.5392588973045349, "lr": 3e-05, "grad_norm": 0.3052245783787724, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121225217783, "tokens_per_second": 149774.38077962978, "data_wait_s": 1.224000152433291e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 828300, "loss": 0.5455509424209595, "lr": 3e-05, "grad_norm": 0.2858742443707635, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121239761221, "tokens_per_second": 150085.2846416131, "data_wait_s": 1.0996001947205514e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 828400, "loss": 0.5238282084465027, "lr": 3e-05, "grad_norm": 0.29655192923119195, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121254383670, "tokens_per_second": 151297.49738241534, "data_wait_s": 1.3859997125109658e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5d6c497c2290e3db4ac3d56f06c74b24694a96b57b530390e4e27aee8824a3f1
-size 603474
+oid sha256:243c220f40f696e60f84522dde792425fa6692e95096dbbe1bd07753d1299471
+size 605592