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": 75145.32974296786,
-      "required_pages_per_second_upper_estimate": 0.006513406728023437
+      "required_token_positions_per_second": 75191.32460007658,
+      "required_pages_per_second_upper_estimate": 0.006517393445664721
     },
     {
       "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": 9393.166217870983,
-      "required_pages_per_second_upper_estimate": 0.0037470744446589208
+      "required_token_positions_per_second": 9398.915575009572,
+      "required_pages_per_second_upper_estimate": 0.0037493679491820537
     },
     {
       "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": 18786.332435741966,
-      "required_pages_per_second_upper_estimate": 0.004345887380850388
+      "required_token_positions_per_second": 18797.831150019145,
+      "required_pages_per_second_upper_estimate": 0.004348547406027955
     },
     {
       "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": 28179.49865361295,
-      "required_pages_per_second_upper_estimate": 0.0017159836216017652
+      "required_token_positions_per_second": 28196.746725028715,
+      "required_pages_per_second_upper_estimate": 0.0017170339386573484
     },
     {
       "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": 28179.49865361295,
-      "required_pages_per_second_upper_estimate": 0.0021693358121999944
+      "required_token_positions_per_second": 28196.746725028715,
+      "required_pages_per_second_upper_estimate": 0.0021706636164832973
     },
     {
       "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": 28179.49865361295,
-      "required_pages_per_second_upper_estimate": 0.3498950625627097
+      "required_token_positions_per_second": 28196.746725028715,
+      "required_pages_per_second_upper_estimate": 0.35010922588411186
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.12469328801572033,
-    "producer_p95_seconds": 0.3487381720005942,
-    "gpu_update_seconds": 0.9811388179702478,
-    "producer_batches_per_second": 8.019677850454373,
-    "gpu_batches_per_second": 1.0192237649599591,
-    "producer_headroom": 7.868417246696981,
+    "producer_batch_seconds": 0.0753448921952895,
+    "producer_p95_seconds": 0.3205180320001091,
+    "gpu_update_seconds": 0.980538651129507,
+    "producer_batches_per_second": 13.272299831660243,
+    "gpu_batches_per_second": 1.0198476101355871,
+    "producer_headroom": 13.014002974322517,
     "sustainable": true,
-    "buffer_seconds": 60.830606714155365,
+    "buffer_seconds": 60.793396370029434,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 383750.43674400245,
+    "generation_bytes_per_second": 383405.01306295313,
     "upload_bytes_per_second": 5542489.025508837,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.34391047229819,
+    "estimated_full_shards_per_hour": 82.26979059139677,
     "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": [

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

@@ -237,3 +237,4 @@
 {"step": 178700, "loss": 0.7068536318838596, "lr": 0.0003, "grad_norm": 0.21037761357928136, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26153880135, "tokens_per_second": 147876.11067979186, "data_wait_s": 8.53799792821519e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 178800, "loss": 0.608285553753376, "lr": 0.0003, "grad_norm": 0.13678190793800188, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26168520724, "tokens_per_second": 149221.04445651054, "data_wait_s": 1.65250021382235e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 178900, "loss": 0.6120023466646671, "lr": 0.0003, "grad_norm": 0.14356371499744083, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26183277476, "tokens_per_second": 149999.27281708477, "data_wait_s": 1.4442997780861333e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 179000, "loss": 0.5916555430740118, "lr": 0.0003, "grad_norm": 0.16379721246225903, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26197937264, "tokens_per_second": 149613.92458736472, "data_wait_s": 1.4415996702155098e-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:dcc326b27aacbdb6d18d471e84119319226377d0be44f7a2e9bd7c11f9d46a51
-size 509499
+oid sha256:481dee9975a69f022d6cb6b75be8cb928efef7fd51ce285474a9cdf0f685c280
+size 511617