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": 75682.13299016481,
-      "required_pages_per_second_upper_estimate": 0.006559935472975077
+      "required_token_positions_per_second": 75190.49737853388,
+      "required_pages_per_second_upper_estimate": 0.006517321744197974
     },
     {
       "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": 9460.266623770602,
-      "required_pages_per_second_upper_estimate": 0.00377384179981275
+      "required_token_positions_per_second": 9398.812172316735,
+      "required_pages_per_second_upper_estimate": 0.003749326700301873
     },
     {
       "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": 18920.533247541203,
-      "required_pages_per_second_upper_estimate": 0.004376932376806522
+      "required_token_positions_per_second": 18797.62434463347,
+      "required_pages_per_second_upper_estimate": 0.004348499565241629
     },
     {
       "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": 28380.799871311803,
-      "required_pages_per_second_upper_estimate": 0.0017282418096137594
+      "required_token_positions_per_second": 28196.436516950205,
+      "required_pages_per_second_upper_estimate": 0.001717015048612903
     },
     {
       "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": 28380.799871311803,
-      "required_pages_per_second_upper_estimate": 0.0021848325371758874
+      "required_token_positions_per_second": 28196.436516950205,
+      "required_pages_per_second_upper_estimate": 0.0021706397358067027
     },
     {
       "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": 28380.799871311803,
-      "required_pages_per_second_upper_estimate": 0.35239454997469244
+      "required_token_positions_per_second": 28196.436516950205,
+      "required_pages_per_second_upper_estimate": 0.3501053741379764
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.08517284599986397,
-    "producer_p95_seconds": 0.3263877190001949,
-    "gpu_update_seconds": 0.9741797315567366,
-    "producer_batches_per_second": 11.740831109501697,
-    "gpu_batches_per_second": 1.0265046249751086,
-    "producer_headroom": 11.437679698507345,
+    "producer_batch_seconds": 0.13531085157029565,
+    "producer_p95_seconds": 0.3462274580015219,
+    "gpu_update_seconds": 0.9805494386987337,
+    "producer_batches_per_second": 7.390390263566464,
+    "gpu_batches_per_second": 1.0198363902253402,
+    "producer_headroom": 7.2466430247046825,
     "sustainable": true,
-    "buffer_seconds": 60.39914335651767,
+    "buffer_seconds": 60.79406519932149,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 331535.26427497296,
+    "generation_bytes_per_second": 332780.77056035644,
     "upload_bytes_per_second": 5059672.6514437925,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.13974996745006,
+    "estimated_full_shards_per_hour": 71.40700662239094,
     "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-v1/1790782192342215557-281f6532/train.jsonl

@@ -320,3 +320,4 @@
 {"step": 387000, "loss": 0.5514918360859156, "lr": 0.0003, "grad_norm": 0.1510423615634618, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56646036656, "tokens_per_second": 146850.150214591, "data_wait_s": 9.527997463010252e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 387100, "loss": 0.5839230827987194, "lr": 0.0003, "grad_norm": 0.17005130184964223, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56660762388, "tokens_per_second": 150913.89009359706, "data_wait_s": 1.0976000339724123e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 387200, "loss": 0.5735466405749321, "lr": 0.0003, "grad_norm": 0.14504306351229077, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56675586356, "tokens_per_second": 150590.1287857146, "data_wait_s": 1.0994001058861613e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 387300, "loss": 0.6117709428071976, "lr": 0.0003, "grad_norm": 0.15717346452028566, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56690278059, "tokens_per_second": 149673.56016251608, "data_wait_s": 1.1883999832207337e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:deb8ee7dd323ab86615922fc9e81a7e02dcca87df67f635c29c41e39ffb64bf0
-size 686859
+oid sha256:5fd1ca36445325dcba1454b896c00989aadb08ad9d1f81fa9c8dd3740b514a8c
+size 688977