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": 77109.73697222857,
-      "required_pages_per_second_upper_estimate": 0.005334630522978809
+      "required_token_positions_per_second": 77063.78524909608,
+      "required_pages_per_second_upper_estimate": 0.005331451476149816
     },
     {
       "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": 9638.71712152857,
-      "required_pages_per_second_upper_estimate": 0.004035529579482283
+      "required_token_positions_per_second": 9632.97315613701,
+      "required_pages_per_second_upper_estimate": 0.004033124701120473
     },
     {
       "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": 19277.43424305714,
-      "required_pages_per_second_upper_estimate": 0.00401498117270073
+      "required_token_positions_per_second": 19265.94631227402,
+      "required_pages_per_second_upper_estimate": 0.004012588539675714
     },
     {
       "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": 28916.151364585712,
-      "required_pages_per_second_upper_estimate": 0.001733479121900287
+      "required_token_positions_per_second": 28898.919468411026,
+      "required_pages_per_second_upper_estimate": 0.0017324460960361965
     },
     {
       "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": 28916.151364585712,
-      "required_pages_per_second_upper_estimate": 0.0022481035798884033
+      "required_token_positions_per_second": 28898.919468411026,
+      "required_pages_per_second_upper_estimate": 0.0022467638757559224
     },
     {
       "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": 28916.151364585712,
-      "required_pages_per_second_upper_estimate": 0.3709053419605407
+      "required_token_positions_per_second": 28898.919468411026,
+      "required_pages_per_second_upper_estimate": 0.3706843096985804
     }
   ],
   "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": 0.0798865070780721,
-    "producer_p95_seconds": 0.1965784240001085,
-    "gpu_update_seconds": 0.9561438398700994,
-    "producer_batches_per_second": 12.517758462298424,
-    "gpu_batches_per_second": 1.045867743221416,
-    "producer_headroom": 11.968777642708448,
+    "producer_batch_seconds": 0.49434742607823523,
+    "producer_p95_seconds": 0.20334100300169666,
+    "gpu_update_seconds": 0.9567139709227402,
+    "producer_batches_per_second": 2.0228688312048386,
+    "gpu_batches_per_second": 1.0452444830877832,
+    "producer_headroom": 1.9353068721578235,
     "sustainable": true,
-    "buffer_seconds": 60.23706191181626,
+    "buffer_seconds": 60.272980168132634,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 89,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 300036.2807995202,
-    "upload_bytes_per_second": 5565945.85868577,
+    "generation_bytes_per_second": 271635.0122302083,
+    "upload_bytes_per_second": 5494465.20072179,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.38080137242513,
+    "estimated_full_shards_per_hour": 58.28655028514564,
     "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": 1.1340931629997613,
-      "bytes": 11786170,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 0.9738086980005392,
       "bytes": 6055828,
@@ -562,6 +556,12 @@
       "bytes": 1149523,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.0760069800016936,
+      "bytes": 5947734,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -114,3 +114,4 @@
 {"step": 615800, "loss": 0.5599251985549927, "lr": 0.0001, "grad_norm": 0.14755198255205235, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90135086258, "tokens_per_second": 150782.9114968828, "data_wait_s": 1.1875999916810542e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 615900, "loss": 0.5587576627731323, "lr": 0.0001, "grad_norm": 0.15822508557198198, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90149703377, "tokens_per_second": 152150.8785774947, "data_wait_s": 1.4863999240333214e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 616000, "loss": 0.5526002645492554, "lr": 0.0001, "grad_norm": 0.1583943054717085, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90164240745, "tokens_per_second": 151031.268838026, "data_wait_s": 1.1617999916779809e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 616100, "loss": 0.5552738308906555, "lr": 0.0001, "grad_norm": 0.15385304505385142, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90178856319, "tokens_per_second": 124564.46547175203, "data_wait_s": 1.091599915525876e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3a3c7b4fd58e6e75385797530d3804259a00f766331d5f4e8e4ce13032e02206
-size 248202
+oid sha256:7229d609c89188b8f128fb2f5553effec59bbc42eee8c9b75795708b53d80c62
+size 250320