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": 75846.90725621907,
-      "required_pages_per_second_upper_estimate": 0.005247265033056632
+      "required_token_positions_per_second": 75890.75043543837,
+      "required_pages_per_second_upper_estimate": 0.005250298205925215
     },
     {
       "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": 9480.863407027384,
-      "required_pages_per_second_upper_estimate": 0.003969439525581036
+      "required_token_positions_per_second": 9486.343804429796,
+      "required_pages_per_second_upper_estimate": 0.003971734053529715
     },
     {
       "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": 18961.726814054768,
-      "required_pages_per_second_upper_estimate": 0.003949227641004322
+      "required_token_positions_per_second": 18972.687608859593,
+      "required_pages_per_second_upper_estimate": 0.00395151048550657
     },
     {
       "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": 28442.59022108215,
-      "required_pages_per_second_upper_estimate": 0.0017050898544332467
+      "required_token_positions_per_second": 28459.03141328939,
+      "required_pages_per_second_upper_estimate": 0.0017060754788018237
     },
     {
       "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": 28442.59022108215,
-      "required_pages_per_second_upper_estimate": 0.002211286283956336
+      "required_token_positions_per_second": 28459.03141328939,
+      "required_pages_per_second_upper_estimate": 0.002212564514333287
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28442.59022108215,
-      "required_pages_per_second_upper_estimate": 0.3648310080820173
+      "required_token_positions_per_second": 28459.03141328939,
+      "required_pages_per_second_upper_estimate": 0.36504189804247494
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.10517302411713558,
-    "producer_p95_seconds": 0.19456041600096796,
-    "gpu_update_seconds": 0.9720633664196594,
-    "producer_batches_per_second": 9.508141544796299,
-    "gpu_batches_per_second": 1.0287395189916866,
-    "producer_headroom": 9.242516078429311,
+    "producer_batch_seconds": 0.0639251970858794,
+    "producer_p95_seconds": 0.2013837019985658,
+    "gpu_update_seconds": 0.9715017914168833,
+    "producer_batches_per_second": 15.643283800229261,
+    "gpu_batches_per_second": 1.0293341801681637,
+    "producer_headroom": 15.197478235565436,
     "sustainable": true,
-    "buffer_seconds": 60.26792871801888,
+    "buffer_seconds": 60.233111067846764,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 350492.5717091685,
-    "upload_bytes_per_second": 3867724.786743171,
+    "generation_bytes_per_second": 376191.68154572137,
+    "upload_bytes_per_second": 3066213.7032676586,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.20754683929721,
+    "estimated_full_shards_per_hour": 80.72197756556254,
     "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": [
@@ -484,6 +484,12 @@
       "bytes": 6181304,
       "pages": 41,
       "shards": 2
+    },
+    {
+      "seconds": 25.14463806799904,
+      "bytes": 7664241,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -104,3 +104,4 @@
 {"step": 695400, "loss": 0.5719165205955505, "lr": 8e-05, "grad_norm": 0.2266536921798782, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101787824083, "tokens_per_second": 152644.94743390344, "data_wait_s": 1.2487998901633546e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 695500, "loss": 0.5675442814826965, "lr": 8e-05, "grad_norm": 0.24226173330281586, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101802556344, "tokens_per_second": 153126.22934712077, "data_wait_s": 9.756999133969657e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 695600, "loss": 0.5088210701942444, "lr": 8e-05, "grad_norm": 0.21322358441997738, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101817276722, "tokens_per_second": 153254.87645242005, "data_wait_s": 1.4193999959388748e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 695700, "loss": 0.525488018989563, "lr": 8e-05, "grad_norm": 0.19512356613865714, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101831984681, "tokens_per_second": 153847.54823968897, "data_wait_s": 1.5897998309810646e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e4753182348680ef4aa93cfa3479f74c2116e180953b06793771b37bebb9ae52
-size 226239
+oid sha256:df46045e7bb61aaab6323a1f84edf9058717556ce4e24872bde45edf7db2a80b
+size 228357