Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75719.18978342562,
-      "required_pages_per_second_upper_estimate": 0.00656314746189534
+      "required_token_positions_per_second": 75779.05499216786,
+      "required_pages_per_second_upper_estimate": 0.006568336426462133
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9464.898722928203,
-      "required_pages_per_second_upper_estimate": 0.0037756896134227714
+      "required_token_positions_per_second": 9472.381874020983,
+      "required_pages_per_second_upper_estimate": 0.00377867475427676
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18929.797445856406,
-      "required_pages_per_second_upper_estimate": 0.004379075486042422
+      "required_token_positions_per_second": 18944.763748041965,
+      "required_pages_per_second_upper_estimate": 0.004382537676655129
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28394.69616878461,
-      "required_pages_per_second_upper_estimate": 0.0017290880212216151
+      "required_token_positions_per_second": 28417.145622062948,
+      "required_pages_per_second_upper_estimate": 0.0017304550751430874
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28394.69616878461,
-      "required_pages_per_second_upper_estimate": 0.0021859023126227575
+      "required_token_positions_per_second": 28417.145622062948,
+      "required_pages_per_second_upper_estimate": 0.002187630533680188
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28394.69616878461,
-      "required_pages_per_second_upper_estimate": 0.3525670954813888
+      "required_token_positions_per_second": 28417.145622062948,
+      "required_pages_per_second_upper_estimate": 0.35284584255761886
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.42742903782826147,
-    "producer_p95_seconds": 0.3531454110016057,
-    "gpu_update_seconds": 0.973702970289026,
-    "producer_batches_per_second": 2.3395696396317236,
-    "gpu_batches_per_second": 1.0270072399010637,
-    "producer_headroom": 2.278045907307435,
+    "producer_batch_seconds": 0.11983179325793003,
+    "producer_p95_seconds": 0.3120235220012546,
+    "gpu_update_seconds": 0.972933748086726,
+    "producer_batches_per_second": 8.345030753629514,
+    "gpu_batches_per_second": 1.0278192137609574,
+    "producer_headroom": 8.119162049027759,
     "sustainable": true,
-    "buffer_seconds": 60.36958415791961,
+    "buffer_seconds": 60.32189238137701,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 366801.7372381555,
-    "upload_bytes_per_second": 5538534.513083096,
+    "generation_bytes_per_second": 367146.6873313597,
+    "upload_bytes_per_second": 5447680.112141541,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.70711410387514,
+    "estimated_full_shards_per_hour": 78.78113236384958,
     "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": 1.0225603920007416,
-      "bytes": 7736202,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 1.0140705839985458,
       "bytes": 1147707,
@@ -562,6 +556,12 @@
       "bytes": 10476048,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.3773436970004695,
+      "bytes": 2708180,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -269,3 +269,4 @@
 {"step": 181900, "loss": 0.5988446213304996, "lr": 0.0003, "grad_norm": 0.13855545873119954, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26622431128, "tokens_per_second": 147935.4875856118, "data_wait_s": 1.4346001989906654e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 182000, "loss": 0.5574628598988056, "lr": 0.0003, "grad_norm": 0.13092034611200506, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26636949683, "tokens_per_second": 148221.26669651538, "data_wait_s": 1.5195997548289597e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 182100, "loss": 0.5786740891635418, "lr": 0.0003, "grad_norm": 0.14327585639957502, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26651454268, "tokens_per_second": 147829.50509332068, "data_wait_s": 1.2241998774698004e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 182200, "loss": 0.5785130281001329, "lr": 0.0003, "grad_norm": 0.1350780864680985, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26666172218, "tokens_per_second": 149499.71403536998, "data_wait_s": 1.3234002835815772e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ef1ff28d86969352096f167737f82ba4bbc882efe289ba37521a5f0a379d2769
-size 578058
+oid sha256:72747db5ce1f4e7fd8da73c126803cd6cabd026b14cee1b60e87ddac3d953bb7
+size 580176