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": 76838.17608953957,
-      "required_pages_per_second_upper_estimate": 0.005315843311006324
+      "required_token_positions_per_second": 76567.33905992791,
+      "required_pages_per_second_upper_estimate": 0.0052971061768693
     },
     {
       "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": 9604.772011192446,
-      "required_pages_per_second_upper_estimate": 0.004021317470639058
+      "required_token_positions_per_second": 9570.917382490989,
+      "required_pages_per_second_upper_estimate": 0.004007143244566796
     },
     {
       "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": 19209.544022384893,
-      "required_pages_per_second_upper_estimate": 0.004000841430120218
+      "required_token_positions_per_second": 19141.834764981977,
+      "required_pages_per_second_upper_estimate": 0.003986739377416386
     },
     {
       "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": 28814.31603357734,
-      "required_pages_per_second_upper_estimate": 0.001727374249299882
+      "required_token_positions_per_second": 28712.752147472966,
+      "required_pages_per_second_upper_estimate": 0.0017212856494069997
     },
     {
       "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": 28814.31603357734,
-      "required_pages_per_second_upper_estimate": 0.0022401863308287853
+      "required_token_positions_per_second": 28712.752147472966,
+      "required_pages_per_second_upper_estimate": 0.002232290185416493
     },
     {
       "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": 28814.31603357734,
-      "required_pages_per_second_upper_estimate": 0.36959910767662474
+      "required_token_positions_per_second": 28712.752147472966,
+      "required_pages_per_second_upper_estimate": 0.3682963551964824
     }
   ],
   "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": 0.10139420798438437,
-    "producer_p95_seconds": 0.30508705700049177,
-    "gpu_update_seconds": 0.9595230359721802,
-    "producer_batches_per_second": 9.862496289275311,
-    "gpu_batches_per_second": 1.0421844630200137,
-    "producer_headroom": 9.46329238174981,
+    "producer_batch_seconds": 0.08247454904690699,
+    "producer_p95_seconds": 0.2957663870001852,
+    "gpu_update_seconds": 0.9629170989250964,
+    "producer_batches_per_second": 12.124952625460917,
+    "gpu_batches_per_second": 1.0385110007043175,
+    "producer_headroom": 11.675324206713055,
     "sustainable": true,
-    "buffer_seconds": 60.44995126624735,
+    "buffer_seconds": 60.66377723228107,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 336377.95343496953,
-    "upload_bytes_per_second": 5597256.262150535,
+    "generation_bytes_per_second": 334995.3667352742,
+    "upload_bytes_per_second": 5548298.119672106,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.17887832915129,
+    "estimated_full_shards_per_hour": 71.88220740836782,
     "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": 1.201682383001753,
-      "bytes": 8258960,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 1.4597207000006165,
       "bytes": 10914765,
@@ -562,6 +556,12 @@
       "bytes": 7573359,
       "pages": 41,
       "shards": 3
+    },
+    {
+      "seconds": 1.4761009860012564,
+      "bytes": 5975966,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -301,3 +301,4 @@
 {"step": 675100, "loss": 0.5807151794433594, "lr": 0.0001, "grad_norm": 0.1930958039307218, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98816785728, "tokens_per_second": 141621.95821272343, "data_wait_s": 1.3183998817112297e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 675200, "loss": 0.5657216310501099, "lr": 0.0001, "grad_norm": 0.2053365528525142, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98831507844, "tokens_per_second": 152895.0128957328, "data_wait_s": 1.4689998351968825e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 675300, "loss": 0.5566164255142212, "lr": 0.0001, "grad_norm": 0.2146848906602766, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98846120568, "tokens_per_second": 152344.26664891464, "data_wait_s": 1.751099989633076e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 675400, "loss": 0.5043987035751343, "lr": 0.0001, "grad_norm": 0.20687713845321776, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98860863971, "tokens_per_second": 154246.7165421657, "data_wait_s": 1.3676002708962187e-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 sha256:233938f4460d6232dee2ab881c3816a54a70c32176053045eac61bfbd9b047de
-size 646617
+oid sha256:9dbf6d650e53c43755bdbb348549baf840187a650ee922161f19034b015e0bbf
+size 648735