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": 75129.04438503968,
-      "required_pages_per_second_upper_estimate": 0.006511995154473085
+      "required_token_positions_per_second": 75199.08341034055,
+      "required_pages_per_second_upper_estimate": 0.006518065959673877
     },
     {
       "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": 9391.13054812996,
-      "required_pages_per_second_upper_estimate": 0.003746262385563252
+      "required_token_positions_per_second": 9399.885426292569,
+      "required_pages_per_second_upper_estimate": 0.00374975483735941
     },
     {
       "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": 18782.26109625992,
-      "required_pages_per_second_upper_estimate": 0.004344945548114345
+      "required_token_positions_per_second": 18799.770852585138,
+      "required_pages_per_second_upper_estimate": 0.004348996122078998
     },
     {
       "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": 28173.39164438988,
-      "required_pages_per_second_upper_estimate": 0.0017156117367810816
+      "required_token_positions_per_second": 28199.65627887771,
+      "required_pages_per_second_upper_estimate": 0.0017172111152215026
     },
     {
       "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": 28173.39164438988,
-      "required_pages_per_second_upper_estimate": 0.002168865677724719
+      "required_token_positions_per_second": 28199.65627887771,
+      "required_pages_per_second_upper_estimate": 0.0021708876019927536
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28173.39164438988,
-      "required_pages_per_second_upper_estimate": 0.34981923394700426
+      "required_token_positions_per_second": 28199.65627887771,
+      "required_pages_per_second_upper_estimate": 0.350145352805266
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.2821552387499935,
-    "producer_p95_seconds": 0.33179157100312295,
-    "gpu_update_seconds": 0.9813514946648427,
-    "producer_batches_per_second": 3.544148265437879,
-    "gpu_batches_per_second": 1.0190028806564626,
-    "producer_headroom": 3.478055197601272,
+    "producer_batch_seconds": 0.1127774761017406,
+    "producer_p95_seconds": 0.3205159099998127,
+    "gpu_update_seconds": 0.9804374821656634,
+    "producer_batches_per_second": 8.867018792811644,
+    "gpu_batches_per_second": 1.0199528457348708,
+    "producer_headroom": 8.693557579539869,
     "sustainable": true,
-    "buffer_seconds": 60.84379266922025,
+    "buffer_seconds": 60.78712389427113,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 419122.69394637115,
-    "upload_bytes_per_second": 5498166.11376146,
+    "generation_bytes_per_second": 417985.2441775344,
+    "upload_bytes_per_second": 5492459.424778028,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.93397344392157,
+    "estimated_full_shards_per_hour": 89.68990320200467,
     "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": 1.9895669670004281,
-      "bytes": 15602081,
-      "pages": 47,
-      "shards": 4
-    },
     {
       "seconds": 0.9570206669995969,
       "bytes": 2289232,
@@ -562,6 +556,12 @@
       "bytes": 1051160,
       "pages": 32,
       "shards": 1
+    },
+    {
+      "seconds": 1.0477113340020878,
+      "bytes": 9989334,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -239,3 +239,4 @@
 {"step": 178900, "loss": 0.6120023466646671, "lr": 0.0003, "grad_norm": 0.14356371499744083, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26183277476, "tokens_per_second": 149999.27281708477, "data_wait_s": 1.4442997780861333e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 179000, "loss": 0.5916555430740118, "lr": 0.0003, "grad_norm": 0.16379721246225903, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26197937264, "tokens_per_second": 149613.92458736472, "data_wait_s": 1.4415996702155098e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 179100, "loss": 0.5237388759851456, "lr": 0.0003, "grad_norm": 0.15559777964325278, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26212625519, "tokens_per_second": 149030.32106352434, "data_wait_s": 1.207899913424626e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 179200, "loss": 0.6327262502163649, "lr": 0.0003, "grad_norm": 0.15070615208766142, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26227355581, "tokens_per_second": 149891.60798435472, "data_wait_s": 1.9053000869462267e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:38721a218a405e93cbe741db697469ca24ad86dad7dfe5879ce0b944ca497769
-size 513735
+oid sha256:bf3dbe8d1420564c659fac9be4bf68524b3ef149e8f5473933716391c91c1dd4
+size 515853