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": 77168.29278915029,
-      "required_pages_per_second_upper_estimate": 0.0053386815502616675
+      "required_token_positions_per_second": 76975.50108747471,
+      "required_pages_per_second_upper_estimate": 0.005325343773001372
     },
     {
       "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": 9646.036598643786,
-      "required_pages_per_second_upper_estimate": 0.0040385940917023605
+      "required_token_positions_per_second": 9621.937635934339,
+      "required_pages_per_second_upper_estimate": 0.004028504359259482
     },
     {
       "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": 19292.073197287573,
-      "required_pages_per_second_upper_estimate": 0.004018030080811745
+      "required_token_positions_per_second": 19243.875271868677,
+      "required_pages_per_second_upper_estimate": 0.0040079917240117575
     },
     {
       "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": 28938.10979593136,
-      "required_pages_per_second_upper_estimate": 0.0017347954963308752
+      "required_token_positions_per_second": 28865.812907803018,
+      "required_pages_per_second_upper_estimate": 0.0017304614082785375
     },
     {
       "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": 28938.10979593136,
-      "required_pages_per_second_upper_estimate": 0.0022498107513405953
+      "required_token_positions_per_second": 28865.812907803018,
+      "required_pages_per_second_upper_estimate": 0.0022441899862890202
     },
     {
       "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": 28938.10979593136,
-      "required_pages_per_second_upper_estimate": 0.371187001140716
+      "required_token_positions_per_second": 28865.812907803018,
+      "required_pages_per_second_upper_estimate": 0.37025965428615615
     }
   ],
   "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": 0.23938429012494566,
-    "producer_p95_seconds": 0.2290029340001638,
-    "gpu_update_seconds": 0.9554183115265964,
-    "producer_batches_per_second": 4.17738356797789,
-    "gpu_batches_per_second": 1.0466619573181193,
-    "producer_headroom": 3.991148755116385,
+    "producer_batch_seconds": 0.14783706200769586,
+    "producer_p95_seconds": 0.23613182799999777,
+    "gpu_update_seconds": 0.957811238100493,
+    "producer_batches_per_second": 6.764203687624309,
+    "gpu_batches_per_second": 1.0440470525102366,
+    "producer_headroom": 6.4788303088073596,
     "sustainable": true,
-    "buffer_seconds": 60.19135362617557,
+    "buffer_seconds": 60.34210800033106,
     "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": 372119.1062006688,
-    "upload_bytes_per_second": 5196451.587772348,
+    "generation_bytes_per_second": 370226.18944809906,
+    "upload_bytes_per_second": 5278938.93980397,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.8480977012162,
+    "estimated_full_shards_per_hour": 79.4419218309615,
     "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": 1.351263048999499,
-      "bytes": 5825113,
-      "pages": 32,
-      "shards": 2
-    },
     {
       "seconds": 1.289677515999756,
       "bytes": 7353219,
@@ -562,6 +556,12 @@
       "bytes": 6384191,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.3574712270001328,
+      "bytes": 12582265,
+      "pages": 37,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -124,3 +124,4 @@
 {"step": 657400, "loss": 0.5541929006576538, "lr": 0.0001, "grad_norm": 0.22244126197697556, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96225596595, "tokens_per_second": 152390.4617840014, "data_wait_s": 9.798999599297531e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 657500, "loss": 0.5422726273536682, "lr": 0.0001, "grad_norm": 0.184700603723759, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96240272694, "tokens_per_second": 152578.73015433078, "data_wait_s": 9.624998710933141e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 657600, "loss": 0.5721409320831299, "lr": 0.0001, "grad_norm": 0.23453476625534417, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96254934317, "tokens_per_second": 153055.72192716855, "data_wait_s": 9.101000614464283e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 657700, "loss": 0.576458215713501, "lr": 0.0001, "grad_norm": 0.20272162505566665, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96269581173, "tokens_per_second": 152844.57654697844, "data_wait_s": 1.2495000191847794e-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 sha256:80ebc494c13faa1e1e27461518adc6d788821baa0b5840938e4dedfa858e8949
-size 268599
+oid sha256:be497d5c7cdb091946da0f18939ccd1f409bb0dc79fe59eacda684973956a186
+size 270717