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": 77224.13903378708,
-      "required_pages_per_second_upper_estimate": 0.005342545123046782
+      "required_token_positions_per_second": 77211.76841455723,
+      "required_pages_per_second_upper_estimate": 0.0053416892949046686
     },
     {
       "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": 9653.017379223385,
-      "required_pages_per_second_upper_estimate": 0.004041516798755763
+      "required_token_positions_per_second": 9651.471051819653,
+      "required_pages_per_second_upper_estimate": 0.004040869383762809
     },
     {
       "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": 19306.03475844677,
-      "required_pages_per_second_upper_estimate": 0.0040209379058100785
+      "required_token_positions_per_second": 19302.942103639307,
+      "required_pages_per_second_upper_estimate": 0.004020293787372476
     },
     {
       "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": 28959.052137670154,
-      "required_pages_per_second_upper_estimate": 0.0017360509577410118
+      "required_token_positions_per_second": 28954.413155458962,
+      "required_pages_per_second_upper_estimate": 0.0017357728578407665
     },
     {
       "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": 28959.052137670154,
-      "required_pages_per_second_upper_estimate": 0.0022514389262951605
+      "required_token_positions_per_second": 28954.413155458962,
+      "required_pages_per_second_upper_estimate": 0.0022510782658329664
     },
     {
       "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": 28959.052137670154,
-      "required_pages_per_second_upper_estimate": 0.3714556270144066
+      "required_token_positions_per_second": 28954.413155458962,
+      "required_pages_per_second_upper_estimate": 0.37139612313155246
     }
   ],
   "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": 0.320315975382897,
-    "producer_p95_seconds": 0.3038523830000486,
-    "gpu_update_seconds": 0.9547273808742957,
-    "producer_batches_per_second": 3.121917346784303,
-    "gpu_batches_per_second": 1.0474194204886487,
-    "producer_headroom": 2.980579971801408,
+    "producer_batch_seconds": 0.31841975050005544,
+    "producer_p95_seconds": 0.28945325399763533,
+    "gpu_update_seconds": 0.954880344200219,
+    "producer_batches_per_second": 3.140508710372304,
+    "gpu_batches_per_second": 1.047251633226959,
+    "producer_headroom": 2.9988100383240917,
     "sustainable": true,
-    "buffer_seconds": 60.147824995080626,
+    "buffer_seconds": 60.15746168461379,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 460519.9288050802,
-    "upload_bytes_per_second": 5253761.760925359,
+    "generation_bytes_per_second": 411934.6118798999,
+    "upload_bytes_per_second": 5409450.973229825,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.81685636629395,
+    "estimated_full_shards_per_hour": 88.39157836244343,
     "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": 1.0295740060028038,
-      "bytes": 1098382,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.3453053759985778,
       "bytes": 7563007,
@@ -562,6 +556,12 @@
       "bytes": 2526136,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 1.477937416999339,
+      "bytes": 16111480,
+      "pages": 34,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -281,3 +281,4 @@
 {"step": 673100, "loss": 0.5786046385765076, "lr": 0.0001, "grad_norm": 0.20524825609338687, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98524328627, "tokens_per_second": 152394.9734899505, "data_wait_s": 1.2312997569097206e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 673200, "loss": 0.5596272945404053, "lr": 0.0001, "grad_norm": 0.19864722771647259, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98538767042, "tokens_per_second": 150574.66379041548, "data_wait_s": 1.32589993881993e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 673300, "loss": 0.5706678628921509, "lr": 0.0001, "grad_norm": 0.19494510413256247, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98553506449, "tokens_per_second": 152480.47147365767, "data_wait_s": 1.3478998880600557e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 673400, "loss": 0.5718695521354675, "lr": 0.0001, "grad_norm": 0.23013544544482514, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98568308512, "tokens_per_second": 154093.95665601574, "data_wait_s": 1.2325999705353752e-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 sha256:917ac2009be1d10685a47c5b6c313ac4f7e5df046bd772050aba19c3566311cd
-size 603474
+oid sha256:27de6d8e4d1420f6b167ce7f709c41af49a87cd51657c2f63bf452e893471f3c
+size 605592