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": 76830.6679569055,
-      "required_pages_per_second_upper_estimate": 0.005315323880969431
+      "required_token_positions_per_second": 76433.66278932882,
+      "required_pages_per_second_upper_estimate": 0.005287858142297574
     },
     {
       "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": 9603.833494613187,
-      "required_pages_per_second_upper_estimate": 0.004020924533345777
+      "required_token_positions_per_second": 9554.207848666103,
+      "required_pages_per_second_upper_estimate": 0.004000147311689062
     },
     {
       "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": 19207.666989226374,
-      "required_pages_per_second_upper_estimate": 0.004000450493613992
+      "required_token_positions_per_second": 19108.415697332206,
+      "required_pages_per_second_upper_estimate": 0.003979779066945019
     },
     {
       "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": 28811.500483839558,
-      "required_pages_per_second_upper_estimate": 0.0017272054613921976
+      "required_token_positions_per_second": 28662.62354599831,
+      "required_pages_per_second_upper_estimate": 0.0017182805162905366
     },
     {
       "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": 28811.500483839558,
-      "required_pages_per_second_upper_estimate": 0.002239967434220967
+      "required_token_positions_per_second": 28662.62354599831,
+      "required_pages_per_second_upper_estimate": 0.0022283929071442543
     },
     {
       "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": 28811.500483839558,
-      "required_pages_per_second_upper_estimate": 0.36956299282769023
+      "required_token_positions_per_second": 28662.62354599831,
+      "required_pages_per_second_upper_estimate": 0.36765335932066434
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.09583534235162716,
-    "producer_p95_seconds": 0.1660637830000269,
-    "gpu_update_seconds": 0.9596168035575874,
-    "producer_batches_per_second": 10.434563861951094,
-    "gpu_batches_per_second": 1.0420826274536878,
-    "producer_headroom": 10.013182819723024,
+    "producer_batch_seconds": 0.16813853267191092,
+    "producer_p95_seconds": 0.16981240199947933,
+    "gpu_update_seconds": 0.9646011627522504,
+    "producer_batches_per_second": 5.947476667655368,
+    "gpu_batches_per_second": 1.036697900245888,
+    "producer_headroom": 5.736942909062248,
     "sustainable": true,
-    "buffer_seconds": 60.455858624128005,
+    "buffer_seconds": 60.76987325339177,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 488934.9756293689,
-    "upload_bytes_per_second": 4681575.855958061,
+    "generation_bytes_per_second": 482563.6657023326,
+    "upload_bytes_per_second": 4637665.9347759355,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 104.9140639463501,
+    "estimated_full_shards_per_hour": 103.5469291525124,
     "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": [
@@ -430,6 +430,12 @@
       "bytes": 6302843,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.0353907819999222,
+      "bytes": 2349590,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -49,3 +49,4 @@
 {"step": 402000, "loss": 0.5714942216873169, "lr": 0.0003, "grad_norm": 0.1585314004546547, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58842105808, "tokens_per_second": 152863.27123831736, "data_wait_s": 1.0505999853194226e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 402100, "loss": 0.5505154728889465, "lr": 0.0003, "grad_norm": 0.172768765517561, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58856733724, "tokens_per_second": 151664.14977400075, "data_wait_s": 1.0599999768601265e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 402200, "loss": 0.629511833190918, "lr": 0.0003, "grad_norm": 0.1567020686235959, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58871240622, "tokens_per_second": 151282.05362207777, "data_wait_s": 7.9849996836856e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 402300, "loss": 0.579757571220398, "lr": 0.0003, "grad_norm": 0.14882517023870942, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58885813869, "tokens_per_second": 151106.00416768625, "data_wait_s": 1.170999985333765e-05, "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 sha256:113f445f941af2084739a94d6cd2b8794bde79bfdefedada53990a27c561d4ae
-size 108966
+oid sha256:06702129600c2479959e2c9d283b207da07c4d74f80dfae121314cc8fb017d2d
+size 111084