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": 76721.89329997978,
-      "required_pages_per_second_upper_estimate": 0.005307798597811333
+      "required_token_positions_per_second": 76440.33572844804,
+      "required_pages_per_second_upper_estimate": 0.0052883197917091935
     },
     {
       "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": 9590.236662497473,
-      "required_pages_per_second_upper_estimate": 0.004015231823673069
+      "required_token_positions_per_second": 9555.041966056006,
+      "required_pages_per_second_upper_estimate": 0.004000496539221862
     },
     {
       "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": 19180.473324994946,
-      "required_pages_per_second_upper_estimate": 0.003994786770499737
+      "required_token_positions_per_second": 19110.08393211201,
+      "required_pages_per_second_upper_estimate": 0.003980126516255343
     },
     {
       "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": 28770.709987492417,
-      "required_pages_per_second_upper_estimate": 0.0017247601334196682
+      "required_token_positions_per_second": 28665.12589816802,
+      "required_pages_per_second_upper_estimate": 0.0017184305284822401
     },
     {
       "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": 28770.709987492417,
-      "required_pages_per_second_upper_estimate": 0.002236796152548409
+      "required_token_positions_per_second": 28665.12589816802,
+      "required_pages_per_second_upper_estimate": 0.002228587454018765
     },
     {
       "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": 28770.709987492417,
-      "required_pages_per_second_upper_estimate": 0.3690397761379718
+      "required_token_positions_per_second": 28665.12589816802,
+      "required_pages_per_second_upper_estimate": 0.3676854568074809
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.0925559762185344,
-    "producer_p95_seconds": 0.3347379810002167,
-    "gpu_update_seconds": 0.9609773277065288,
-    "producer_batches_per_second": 10.804272623508338,
-    "gpu_batches_per_second": 1.0406072767466874,
-    "producer_headroom": 10.38266103355185,
+    "producer_batch_seconds": 0.14112461832752388,
+    "producer_p95_seconds": 0.3389008689991897,
+    "gpu_update_seconds": 0.9645169568840783,
+    "producer_batches_per_second": 7.085935904387616,
+    "gpu_batches_per_second": 1.0367884077751743,
+    "producer_headroom": 6.834505335175573,
     "sustainable": true,
-    "buffer_seconds": 60.54157164551131,
+    "buffer_seconds": 60.76456828369694,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 451967.1924105777,
-    "upload_bytes_per_second": 5393423.466757317,
+    "generation_bytes_per_second": 451961.6254880117,
+    "upload_bytes_per_second": 5299547.7499331115,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.9816382335472,
+    "estimated_full_shards_per_hour": 96.98044370155586,
     "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": 1.1926017650002905,
-      "bytes": 13608076,
-      "pages": 35,
-      "shards": 4
-    },
     {
       "seconds": 0.8606232550009736,
       "bytes": 4749519,
@@ -562,6 +556,12 @@
       "bytes": 5389343,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.2195735799978138,
+      "bytes": 6166768,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -356,3 +356,4 @@
 {"step": 432700, "loss": 0.596798837184906, "lr": 0.0003, "grad_norm": 0.15001806380939847, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63335096869, "tokens_per_second": 150830.0575688252, "data_wait_s": 1.727099879644811e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 432800, "loss": 0.6232839822769165, "lr": 0.0003, "grad_norm": 0.15557434976103673, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63349812933, "tokens_per_second": 152483.18746350807, "data_wait_s": 1.3459000911097974e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 432900, "loss": 0.5553624033927917, "lr": 0.0003, "grad_norm": 0.15352128821337596, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63364505407, "tokens_per_second": 151727.28399990578, "data_wait_s": 1.2948003131896257e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 433000, "loss": 0.6017171144485474, "lr": 0.0003, "grad_norm": 0.15152019661519708, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63379170427, "tokens_per_second": 151419.2022181481, "data_wait_s": 1.349899685010314e-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:f309d47120974eb487e9eb982ffcf7f5a3c414902a2f792d33bd6ba5c5aad95c
-size 763890
+oid sha256:48af39bf6baeed27be52fdb4a1b44e70d1ad3fd89aaa5bb27e4f917c09f7ec8c
+size 766008