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": 76427.96792979668,
-      "required_pages_per_second_upper_estimate": 0.005287464158701246
+      "required_token_positions_per_second": 76077.98893022674,
+      "required_pages_per_second_upper_estimate": 0.005263251799447838
     },
     {
       "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": 9553.495991224585,
-      "required_pages_per_second_upper_estimate": 0.003999849271843572
+      "required_token_positions_per_second": 9509.748616278343,
+      "required_pages_per_second_upper_estimate": 0.003981533159502652
     },
     {
       "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": 19106.99198244917,
-      "required_pages_per_second_upper_estimate": 0.003979482544680769
+      "required_token_positions_per_second": 19019.497232556685,
+      "required_pages_per_second_upper_estimate": 0.003961259695669884
     },
     {
       "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": 28660.487973673753,
-      "required_pages_per_second_upper_estimate": 0.0017181524919905094
+      "required_token_positions_per_second": 28529.245848835024,
+      "required_pages_per_second_upper_estimate": 0.001710284727001549
     },
     {
       "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": 28660.487973673753,
-      "required_pages_per_second_upper_estimate": 0.0022282268757894096
+      "required_token_positions_per_second": 28529.245848835024,
+      "required_pages_per_second_upper_estimate": 0.0022180233778562974
     },
     {
       "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": 28660.487973673753,
-      "required_pages_per_second_upper_estimate": 0.3676259664918838
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28529.245848835024,
+      "required_pages_per_second_upper_estimate": 0.3659425334312672
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.12295133571890915,
-    "producer_p95_seconds": 0.29234366300079273,
-    "gpu_update_seconds": 0.9646730378560274,
-    "producer_batches_per_second": 8.133299196408862,
-    "gpu_batches_per_second": 1.0366206587700286,
-    "producer_headroom": 7.845974443591723,
+    "producer_batch_seconds": 0.07935770467992143,
+    "producer_p95_seconds": 0.2882902710007329,
+    "gpu_update_seconds": 0.9691107906074913,
+    "producer_batches_per_second": 12.601170913818196,
+    "gpu_batches_per_second": 1.0318737647871465,
+    "producer_headroom": 12.211930706870476,
     "sustainable": true,
-    "buffer_seconds": 60.774401384929725,
+    "buffer_seconds": 60.08486901766446,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 438323.17368805583,
-    "upload_bytes_per_second": 6002318.867785878,
+    "generation_bytes_per_second": 420691.19295673765,
+    "upload_bytes_per_second": 5894810.864061755,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.05394943219429,
+    "estimated_full_shards_per_hour": 90.27053681875799,
     "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": 1.2385705720007536,
-      "bytes": 15798086,
-      "pages": 39,
-      "shards": 4
-    },
     {
       "seconds": 0.9273680320002313,
       "bytes": 2684614,
@@ -562,6 +556,12 @@
       "bytes": 10696457,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.1242637430004834,
+      "bytes": 7345030,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -273,3 +273,4 @@
 {"step": 424400, "loss": 0.6360502243041992, "lr": 0.0003, "grad_norm": 0.16612847388856186, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62120496160, "tokens_per_second": 150041.53417072596, "data_wait_s": 1.3700999261345714e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 424500, "loss": 0.5506383776664734, "lr": 0.0003, "grad_norm": 0.184131974403261, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62135124919, "tokens_per_second": 151682.1637055532, "data_wait_s": 1.3084001693641767e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 424600, "loss": 0.5884664058685303, "lr": 0.0003, "grad_norm": 0.1666712249518507, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62149745642, "tokens_per_second": 150856.40673508198, "data_wait_s": 1.9003000488737598e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 424700, "loss": 0.589414656162262, "lr": 0.0003, "grad_norm": 0.15922639864515833, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62164424757, "tokens_per_second": 152461.7311451928, "data_wait_s": 8.151997462846339e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f83605a8a07d88ae133d8fb380591610cfe1c71491a5f191ee752ae259deb2df
-size 586530
+oid sha256:80ec2fe77965a9f081c92c7d7d4a746fb47c37a55c63d499daa4ab517415b253
+size 588648