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": 76864.35545607637,
-      "required_pages_per_second_upper_estimate": 0.005317654460327839
+      "required_token_positions_per_second": 76161.75118539596,
+      "required_pages_per_second_upper_estimate": 0.005269046666615654
     },
     {
       "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": 9608.044432009547,
-      "required_pages_per_second_upper_estimate": 0.00402268756489926
+      "required_token_positions_per_second": 9520.218898174495,
+      "required_pages_per_second_upper_estimate": 0.003985916847888222
     },
     {
       "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": 19216.088864019093,
-      "required_pages_per_second_upper_estimate": 0.004002204548033542
+      "required_token_positions_per_second": 19040.43779634899,
+      "required_pages_per_second_upper_estimate": 0.003965621062868017
     },
     {
       "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": 28824.13329602864,
-      "required_pages_per_second_upper_estimate": 0.001727962779193746
+      "required_token_positions_per_second": 28560.656694523484,
+      "required_pages_per_second_upper_estimate": 0.001712167759238988
     },
     {
       "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": 28824.13329602864,
-      "required_pages_per_second_upper_estimate": 0.0022409495798028
+      "required_token_positions_per_second": 28560.656694523484,
+      "required_pages_per_second_upper_estimate": 0.0022204654329468663
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28824.13329602864,
-      "required_pages_per_second_upper_estimate": 0.3697250329783949
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28560.656694523484,
+      "required_pages_per_second_upper_estimate": 0.36634543803342035
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10319414148409578,
-    "producer_p95_seconds": 0.30648944999848027,
-    "gpu_update_seconds": 0.9591962303272207,
-    "producer_batches_per_second": 9.69047259484318,
-    "gpu_batches_per_second": 1.0425395434038138,
-    "producer_headroom": 9.29506478306282,
+    "producer_batch_seconds": 0.08912029598457138,
+    "producer_p95_seconds": 0.3109928650010261,
+    "gpu_update_seconds": 0.9680449681432399,
+    "producer_batches_per_second": 11.220788586396989,
+    "gpu_batches_per_second": 1.03300986308317,
+    "producer_headroom": 10.862227929660703,
     "sustainable": true,
-    "buffer_seconds": 60.4293625106149,
+    "buffer_seconds": 60.01878802488088,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 317791.22316150245,
-    "upload_bytes_per_second": 5586075.304059285,
+    "generation_bytes_per_second": 347085.38511931244,
+    "upload_bytes_per_second": 5546548.94052357,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.19059868940168,
+    "estimated_full_shards_per_hour": 74.47644391235858,
     "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": 1.2478048169978138,
-      "bytes": 10304832,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 0.9087454849977803,
       "bytes": 2381627,
@@ -562,6 +556,12 @@
       "bytes": 6396203,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.070789342003991,
+      "bytes": 6236244,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -336,3 +336,4 @@
 {"step": 678600, "loss": 0.5879587531089783, "lr": 0.0001, "grad_norm": 0.19140846977270953, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99329593942, "tokens_per_second": 150517.2255848945, "data_wait_s": 9.38000448513776e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 678700, "loss": 0.48311981558799744, "lr": 0.0001, "grad_norm": 0.2524622547476496, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99344159402, "tokens_per_second": 152253.01380390886, "data_wait_s": 1.0231997293885797e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 678800, "loss": 0.569308876991272, "lr": 0.0001, "grad_norm": 0.20589001665957765, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99358795489, "tokens_per_second": 151683.22996427878, "data_wait_s": 1.487800182076171e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 678900, "loss": 0.6020879149436951, "lr": 0.0001, "grad_norm": 0.2556691093215879, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99373500312, "tokens_per_second": 152638.41731959954, "data_wait_s": 1.2704003893304616e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e90e7e9280ae5781f8a719a5ab6f9937d03e22d72045b72c908c68404275b872
-size 720747
+oid sha256:a7bcc006e3845f781925362511a3b18bde05f732c19559c523a63067ab995c13
+size 722865