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": 77097.49140709358,
-      "required_pages_per_second_upper_estimate": 0.005333783346369144
+      "required_token_positions_per_second": 77218.80019012268,
+      "required_pages_per_second_upper_estimate": 0.005342175769454254
     },
     {
       "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": 9637.186425886697,
-      "required_pages_per_second_upper_estimate": 0.00403488870918159
+      "required_token_positions_per_second": 9652.350023765335,
+      "required_pages_per_second_upper_estimate": 0.004041237390961444
     },
     {
       "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": 19274.372851773394,
-      "required_pages_per_second_upper_estimate": 0.004014343565630643
+      "required_token_positions_per_second": 19304.70004753067,
+      "required_pages_per_second_upper_estimate": 0.004020659920724949
     },
     {
       "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": 28911.55927766009,
-      "required_pages_per_second_upper_estimate": 0.0017332038333008066
+      "required_token_positions_per_second": 28957.050071296006,
+      "required_pages_per_second_upper_estimate": 0.0017359309369188599
     },
     {
       "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": 28911.55927766009,
-      "required_pages_per_second_upper_estimate": 0.002247746565328392
+      "required_token_positions_per_second": 28957.050071296006,
+      "required_pages_per_second_upper_estimate": 0.00225128327444073
     },
     {
       "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": 28911.55927766009,
-      "required_pages_per_second_upper_estimate": 0.3708464396000576
+      "required_token_positions_per_second": 28957.050071296006,
+      "required_pages_per_second_upper_estimate": 0.37142994665661044
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.10662828361714105,
-    "producer_p95_seconds": 0.318368919997738,
-    "gpu_update_seconds": 0.9562957063116122,
-    "producer_batches_per_second": 9.37837472457678,
-    "gpu_batches_per_second": 1.045701652114442,
-    "producer_headroom": 8.968499481294124,
+    "producer_batch_seconds": 0.11732634744524262,
+    "producer_p95_seconds": 0.3005195280020416,
+    "gpu_update_seconds": 0.9547933899318835,
+    "producer_batches_per_second": 8.523234736057134,
+    "gpu_batches_per_second": 1.0473470077870373,
+    "producer_headroom": 8.137928186825173,
     "sustainable": true,
-    "buffer_seconds": 60.24662949763157,
+    "buffer_seconds": 60.151983565708655,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 390723.70355555986,
-    "upload_bytes_per_second": 6171697.619300837,
+    "generation_bytes_per_second": 401591.0863426738,
+    "upload_bytes_per_second": 6075840.899860284,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.84021120071503,
+    "estimated_full_shards_per_hour": 86.17209856710588,
     "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": 1.1965479969985608,
-      "bytes": 13159361,
-      "pages": 38,
-      "shards": 4
-    },
     {
       "seconds": 1.0418553449999308,
       "bytes": 11343588,
@@ -562,6 +556,12 @@
       "bytes": 6350229,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.0111840479985403,
+      "bytes": 5239532,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -288,3 +288,4 @@
 {"step": 513800, "loss": 0.6258680820465088, "lr": 0.0003, "grad_norm": 0.16624693792826142, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75202696003, "tokens_per_second": 151591.762574677, "data_wait_s": 1.395299841533415e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 513900, "loss": 0.5715893507003784, "lr": 0.0003, "grad_norm": 0.18317833144933246, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75217419312, "tokens_per_second": 153774.40437191527, "data_wait_s": 9.772997145773843e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 514000, "loss": 0.6018063426017761, "lr": 0.0003, "grad_norm": 0.1565306947475214, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75232211494, "tokens_per_second": 153745.36722023523, "data_wait_s": 1.8021000869339332e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 514100, "loss": 0.5951921343803406, "lr": 0.0003, "grad_norm": 0.17343533276240528, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75247055217, "tokens_per_second": 154014.69632251878, "data_wait_s": 1.586900179972872e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7ae2f6536c53559cefbea59ef4dbbcda3601f70eefe9aaba1134a794d7eef6ab
-size 618300
+oid sha256:950568e6dbcec2f43b9c66b93e63ebe9c657ea91054089548ed9292d21a5939a
+size 620418