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": 76668.71420286849,
-      "required_pages_per_second_upper_estimate": 0.005304119544480668
+      "required_token_positions_per_second": 76591.24814385294,
+      "required_pages_per_second_upper_estimate": 0.005298760262771951
     },
     {
       "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": 9583.589275358561,
-      "required_pages_per_second_upper_estimate": 0.004012448701491235
+      "required_token_positions_per_second": 9573.906017981617,
+      "required_pages_per_second_upper_estimate": 0.004008394523836917
     },
     {
       "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": 19167.178550717123,
-      "required_pages_per_second_upper_estimate": 0.003992017819624442
+      "required_token_positions_per_second": 19147.812035963234,
+      "required_pages_per_second_upper_estimate": 0.003987984285330546
     },
     {
       "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": 28750.767826075684,
-      "required_pages_per_second_upper_estimate": 0.0017235646312926524
+      "required_token_positions_per_second": 28721.718053944853,
+      "required_pages_per_second_upper_estimate": 0.0017218231418098427
     },
     {
       "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": 28750.767826075684,
-      "required_pages_per_second_upper_estimate": 0.0022352457372145557
+      "required_token_positions_per_second": 28721.718053944853,
+      "required_pages_per_second_upper_estimate": 0.002232987245207827
     },
     {
       "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": 28750.767826075684,
-      "required_pages_per_second_upper_estimate": 0.3687839795035426
+      "required_token_positions_per_second": 28721.718053944853,
+      "required_pages_per_second_upper_estimate": 0.36841136021786347
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.10526186014067207,
-    "producer_p95_seconds": 0.27886092599874246,
-    "gpu_update_seconds": 0.9616438826000493,
-    "producer_batches_per_second": 9.500117123748325,
-    "gpu_batches_per_second": 1.039885989079705,
-    "producer_headroom": 9.135729516036552,
+    "producer_batch_seconds": 0.09385988902351983,
+    "producer_p95_seconds": 0.2696618560003117,
+    "gpu_update_seconds": 0.9626165101987212,
+    "producer_batches_per_second": 10.654178375913226,
+    "gpu_batches_per_second": 1.0388352884094638,
+    "producer_headroom": 10.255888007256267,
     "sustainable": true,
-    "buffer_seconds": 60.583564603803104,
+    "buffer_seconds": 60.644840142519435,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 357854.62686666014,
-    "upload_bytes_per_second": 4807959.195163555,
+    "generation_bytes_per_second": 358477.1898213598,
+    "upload_bytes_per_second": 4812927.239782169,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.78727249622204,
+    "estimated_full_shards_per_hour": 76.92086001377675,
     "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": 1.4908248419997108,
-      "bytes": 6812358,
-      "pages": 30,
-      "shards": 2
-    },
     {
       "seconds": 1.521788887999719,
       "bytes": 7493690,
@@ -562,6 +556,12 @@
       "bytes": 7493698,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.2196332230014377,
+      "bytes": 5954986,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -248,3 +248,4 @@
 {"step": 949800, "loss": 0.5232121348381042, "lr": 3e-05, "grad_norm": 0.3377251694705753, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139024570477, "tokens_per_second": 153855.84578278093, "data_wait_s": 1.2970998795935884e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 949900, "loss": 0.501677393913269, "lr": 3e-05, "grad_norm": 0.3597021263595604, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139039255724, "tokens_per_second": 151817.9165736854, "data_wait_s": 1.2939999578520656e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 950000, "loss": 0.5496968626976013, "lr": 3e-05, "grad_norm": 0.3429020703971297, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139053816741, "tokens_per_second": 151462.30566624465, "data_wait_s": 1.371200050925836e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 950100, "loss": 0.5853739976882935, "lr": 3e-05, "grad_norm": 0.35853561981526255, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139068560747, "tokens_per_second": 145432.7120355331, "data_wait_s": 1.4164997992338613e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d127548c23aac9b752186603ecede67a89f3c5c042b4ea29bd453020c377cd93
-size 533580
+oid sha256:c1c7582f93ef212cc68b6c60bafd2da347bd632b0c3b315ca17350ec248e9e02
+size 535698