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": 76441.90771957701,
-      "required_pages_per_second_upper_estimate": 0.005288428545703017
+      "required_token_positions_per_second": 76676.06822472718,
+      "required_pages_per_second_upper_estimate": 0.005304628312776534
     },
     {
       "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": 9555.238464947126,
-      "required_pages_per_second_upper_estimate": 0.004000578809204211
+      "required_token_positions_per_second": 9584.508528090897,
+      "required_pages_per_second_upper_estimate": 0.004012833573414084
     },
     {
       "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": 19110.476929894252,
-      "required_pages_per_second_upper_estimate": 0.003980208367329334
+      "required_token_positions_per_second": 19169.017056181794,
+      "required_pages_per_second_upper_estimate": 0.003992400731828083
     },
     {
       "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": 28665.71539484138,
-      "required_pages_per_second_upper_estimate": 0.0017184658679076988
+      "required_token_positions_per_second": 28753.52558427269,
+      "required_pages_per_second_upper_estimate": 0.00172372995468571
     },
     {
       "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": 28665.71539484138,
-      "required_pages_per_second_upper_estimate": 0.002228633284792197
+      "required_token_positions_per_second": 28753.52558427269,
+      "required_pages_per_second_upper_estimate": 0.002235460140784277
     },
     {
       "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": 28665.71539484138,
-      "required_pages_per_second_upper_estimate": 0.36769301823785455
+      "required_token_positions_per_second": 28753.52558427269,
+      "required_pages_per_second_upper_estimate": 0.36881935306464375
     }
   ],
   "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": 0.2586055518282251,
-    "producer_p95_seconds": 0.33536504399671685,
-    "gpu_update_seconds": 0.9644971220559692,
-    "producer_batches_per_second": 3.8668930072477146,
-    "gpu_batches_per_second": 1.0368097292694365,
-    "producer_headroom": 3.7296071767887726,
+    "producer_batch_seconds": 0.11085214392966236,
+    "producer_p95_seconds": 0.35196908799844095,
+    "gpu_update_seconds": 0.9615516510824892,
+    "producer_batches_per_second": 9.021025345567674,
+    "gpu_batches_per_second": 1.0399857343848629,
+    "producer_headroom": 8.67418181548758,
     "sustainable": true,
-    "buffer_seconds": 60.76331868952606,
+    "buffer_seconds": 60.57775401819682,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 431674.1142942815,
-    "upload_bytes_per_second": 5402553.786746793,
+    "generation_bytes_per_second": 450087.0895735657,
+    "upload_bytes_per_second": 5364126.784592917,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.62721606847128,
+    "estimated_full_shards_per_hour": 96.5782119312785,
     "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": 0.9211728519985627,
-      "bytes": 7602156,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 0.7880418419990747,
       "bytes": 1256366,
@@ -562,6 +556,12 @@
       "bytes": 13551509,
       "pages": 37,
       "shards": 4
+    },
+    {
+      "seconds": 1.065682766995451,
+      "bytes": 5277203,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -375,3 +375,4 @@
 {"step": 434600, "loss": 0.5344572067260742, "lr": 0.0003, "grad_norm": 0.12966294786555346, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63612569088, "tokens_per_second": 152253.80408577007, "data_wait_s": 1.5035999240353703e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 434700, "loss": 0.49582839012145996, "lr": 0.0003, "grad_norm": 0.15662716729449958, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63627304425, "tokens_per_second": 150859.48167857932, "data_wait_s": 1.2040000001434237e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 434800, "loss": 0.593290388584137, "lr": 0.0003, "grad_norm": 0.1732578900896293, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63642137219, "tokens_per_second": 154427.10465974387, "data_wait_s": 1.789099769666791e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 434900, "loss": 0.5564521551132202, "lr": 0.0003, "grad_norm": 0.16359301725097797, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63656864217, "tokens_per_second": 152202.06379904406, "data_wait_s": 1.2739001249428838e-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 sha256:0136c1a15168a07a79573853883c30477760ecca517811a59278e6085a9c926a
-size 804132
+oid sha256:f6e9c9d710eec2a08e0485d2894dc2ec31a9db0b407f353ce54d7af9f48d4db3
+size 806250