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": 74186.33000808196,
-      "required_pages_per_second_upper_estimate": 0.0064302830615661334
+      "required_token_positions_per_second": 74909.22586642942,
+      "required_pages_per_second_upper_estimate": 0.006492941842404889
     },
     {
       "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": 9273.291251010245,
-      "required_pages_per_second_upper_estimate": 0.0036992545280876993
+      "required_token_positions_per_second": 9363.653233303678,
+      "required_pages_per_second_upper_estimate": 0.003735301273856581
     },
     {
       "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": 18546.58250202049,
-      "required_pages_per_second_upper_estimate": 0.004290425453138646
+      "required_token_positions_per_second": 18727.306466607355,
+      "required_pages_per_second_upper_estimate": 0.004332232761712673
     },
     {
       "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": 27819.873753030733,
-      "required_pages_per_second_upper_estimate": 0.0016940843519623416
+      "required_token_positions_per_second": 28090.959699911033,
+      "required_pages_per_second_upper_estimate": 0.001710592063849308
     },
     {
       "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": 27819.873753030733,
-      "required_pages_per_second_upper_estimate": 0.002141650891847735
+      "required_token_positions_per_second": 28090.959699911033,
+      "required_pages_per_second_upper_estimate": 0.0021625198384525104
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27819.873753030733,
-      "required_pages_per_second_upper_estimate": 0.34542972488459633
+      "required_token_positions_per_second": 28090.959699911033,
+      "required_pages_per_second_upper_estimate": 0.3487957050785482
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.07039472879691289,
-    "producer_p95_seconds": 0.25866376499834587,
-    "gpu_update_seconds": 0.9938219075127179,
-    "producer_batches_per_second": 14.20560910014976,
-    "gpu_batches_per_second": 1.0062164985905213,
-    "producer_headroom": 14.117845533290858,
+    "producer_batch_seconds": 0.11735742426563434,
+    "producer_p95_seconds": 0.32239856899832375,
+    "gpu_update_seconds": 0.9842312365030221,
+    "producer_batches_per_second": 8.520977741779129,
+    "gpu_batches_per_second": 1.0160214011831248,
+    "producer_headroom": 8.386612459006,
     "sustainable": true,
-    "buffer_seconds": 60.623136358275794,
+    "buffer_seconds": 60.03810542668435,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 439694.7107834561,
-    "upload_bytes_per_second": 5130724.161041566,
+    "generation_bytes_per_second": 440615.05307177803,
+    "upload_bytes_per_second": 4962278.859892928,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.34824936511767,
+    "estimated_full_shards_per_hour": 94.54573339571958,
     "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": 1.1941432809999242,
-      "bytes": 7137236,
-      "pages": 22,
-      "shards": 3
-    },
     {
       "seconds": 1.292083440001079,
       "bytes": 7124472,
@@ -562,6 +556,12 @@
       "bytes": 12087341,
       "pages": 41,
       "shards": 3
+    },
+    {
+      "seconds": 2.7144779629998084,
+      "bytes": 851325,
+      "pages": 25,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -212,3 +212,4 @@
 {"step": 216200, "loss": 0.5665174443274736, "lr": 0.0003, "grad_norm": 0.16151457878316305, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31644759811, "tokens_per_second": 148242.5218466121, "data_wait_s": 6.607002433156595e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 216300, "loss": 0.5820259880274534, "lr": 0.0003, "grad_norm": 0.13927725451570416, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31659407184, "tokens_per_second": 148922.5319761339, "data_wait_s": 9.254999895347282e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 216400, "loss": 0.5731642600148916, "lr": 0.0003, "grad_norm": 0.14235341976407173, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31674067785, "tokens_per_second": 149381.82045872966, "data_wait_s": 9.686998964753002e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 216500, "loss": 0.5662705302238464, "lr": 0.0003, "grad_norm": 0.16423504628066296, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31688855500, "tokens_per_second": 149772.7716992007, "data_wait_s": 1.089499710360542e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e6ce312113ec024ca110114995b480b0d2b631f8255fa36433b642d2a3ec733d
-size 456549
+oid sha256:ebe1ba8ac7b946717bc24c8a8a60b4f16da95351eef31c28308d21382b57a488
+size 458667